Linguistics and Philosophy 24:659-735. 2001.

The Meaning of Free Choice

Anastasia Giannakidou

University of Groningen

Abstract

In this paper, I discuss the distribution and interpretation of free choice items (FCIs) in Greek, a

language exhibiting a lexical paradigm of such items distinct from that of negative polarity

items. Greek differs in this respect from English which uniformly employs any. FCIs are

grammatical only in certain contexts that can be characterized as nonveridical (Giannakidou

1998, 1999), and although they yield universal-like interpretations in certain structures, they are

not, I argue, universal quantifiers. Evidence will be provided that FCIs are indefinites; the quasi-

universal effect is shown to be the result of binding by an operator with universal force.

Additionally, the limited distribution of FCIs in nonveridical contexts can be accounted for by

analyzing them as indefinites which must always be interpreted in an intensional type. The

difference between “regular” indefinites and FCIs, therefore, is reduced to a type difference

which captures the fact that only the latter exhibit limited distribution: because of their

intensional type, FCIs will be grammatical only in contexts providing alternatives (worlds or

situations), and nonveridical contexts do exactly this. By contrast, FCIs are excluded from

veridical and episodic contexts because these provide no alternatives and hence do not satisfy

the lexical semantic requirement of FCIs. The proposed analysis is supported by data from

other languages as well (Spanish, Catalan, French) and has important consequences regarding

the analysis of English any. If FCIs are not universal quantifiers but indefinites, then the usual

ambiguity thesis (free choice any being universal, negative polarity any an existential) can no

longer be maintained, at least not as one in terms of quantificational force.

1 The problem of free choice

Consider a language like English which possesses the notorious item any and employs it in the

two cases below:

(1) a Did Ariadne talk to anybody?

b Anybody can solve this problem.

In the first sentence anybody seems to contribute an existential quantifier, a likely paraphrase of

2

(1a) being Is there an x, such that x is a person and Ariadne talked to x?. In this instance,

anybody has been characterized as ‘negative’ polarity item. 1 In the second sentence, on the

other hand, anybody seems to be interpreted as the universal quantifier: Every person x is such

that x can solve this problem. In this second use any has been characterized as a free choice

item (FCI).

Consider now (2), with the FC any in an imperative:

(2) Press any key to continue.

A likely paraphrase of this sentence renders FC any equivalent to an existential quantifier: You

must press some key; it doesn’t matter which one. You certainly don’t have to press every key

on your keyboard for the program to continue executing. If free choice can be taken to

contribute a universal quantifier in some cases and an existential in others, how do we decide

what it means?

Trying to provide an answer to this question by concentrating on a language like

English, where there is a real possibility that FC any might be mistaken for its API congener, is

a complicated, tricky, and dangerous business. It is akin to undertaking a study of the semantic

differences between definiteness and indefiniteness on the basis of a language like Russian,

which fails to lexicalize these distinctions in articles, or to providing a semantics for

distinguishing simple and progressive aspects by concentrating only on German, which lacks a

morphological distinction between these, or indeed to elucidating the nature of case by

examining only the English pronominal system. While these are not impossible undertakings in

the abstract, they are Sisyphean tasks indeed, and any analyst would be wise to concetrate their

efforts instead on a language where the relevant distinctions might be discerned more easily, and

judgments made more reliable, by the morphological resources of the language in question. Of

course, simply having a lexical distinction that seems on brief inspection to encode some

theoretically significant category is no guarantee of success, but it does have numerous obvious

advantages as an empirical strategy. For the case at hand, I would like to suggest that we

concentrate initially on a language that does seem to lexicalize the distinction between FCIs and

APIs, as giving us a better chance of reaching a reliable picture of what free choice sensu stricto

means. One such language is Greek.

Greek has a paradigm of items that I will call free choice items (FCIs) and which are

morphologically distinct from that of affective polarity items (APIs) such as kanenas and tipota

1 Or more appropriately, 'affective' polarity item (API), following the terminology of Giannakidou 1998, 1999.APIs are polarity items (PIs) which are grammatical in ‘affective’ contexts (a term coined in Klima 1964),questions and negation being among such contexts. The term ‘negative polarity item’ (NPI) is moreappropriately reserved to single out PIs which are only licensed in negative contexts.

3

(Giannakidou 1997, 1998, 1999). The basic paradigm of FCIs is illustrated in (3), which I gloss

with ‘FC’ and translate varyingly as any(one) , wh-ever for convenience:

(3) opjosdhipote FC-person; anyone, anybody, whoever

otidhipote FC-thing; anything, whatever

opotedhipote FC-time; any time, whenever

opudhipote FC-place; any place, wherever

Morphologically, the FCIs can be decomposed into three parts: opjos, a wh-determiner used

also in forming free relatives (distinct from, though closely related to, the interrogative and

headed relative wh-elements), dhi “indeed” (an undeclinable emphatic particle from ancient

Greek), and pote “ever”(cf. the English -ever paradigm).2 Employing a wh-paradigm and some

kind of modal marking seems to be a common strategy for FCI formation crosslinguistically:

most of the FCI-paradigms identified in the literature involve constituency parallel to the one we

observe in Greek FCIs; see Haspelmath 1997 for an overview, Bosque 1996, Quer 1998, 1999

for Spanish and Catalan, Rullmann 1995 for Dutch, Vikner 1999 for Danish; Sæbø (this issue)

for Norwegian and Swedish, and Dayal 1997 for Hindi. Sometimes concessive markers are also

used (e.g. in Korean, Lee 1997), which seem to reflect the presence of scalarity in the free

choice meaning.

In (4a) we see that FCIs and APIs are both ungrammatical in positive sentences, while

differing in their distribution under negation as the contrast between (4b) and (4c) indicates.

(4) a *Idha {kanenan/ opjondhipote}.

saw.perf.1sg API-person / FC-person

‘*I saw anybody.’

b Dhen idha kanenan.

not saw.perf.1sg API-person

‘I didn’t see anybody.’

c *Dhen idha opjondhipote.

not saw.perf.1sg FC-person

(‘I didn’t see anybody.’)

Sentences like the ones in (4) with verbs in the past and with perfective aspect are about exactly

one event, and can be taken to involve an abstract logical structure like that in (5):

2 Opjosdhipote can be used either as an independent DP constituent, or as a determiner, and like all nominalconstituents in Greek is declined for case, number and gender: nom.sg.masc. opjosdhipote; acc.sg.masc.opjondhipote; gen.sg.masc. opjudhipote; etc. I will use these alternations in the examples below.

4

(5) ∃!e φ(e)

I follow the standard practice in calling sentences with this logical structure episodic. The fact

that both APIs and FCIs are excluded from affirmative episodic sentences qualifies them both

for PI-status. On the other hand, (4c) indicates that, unlike APIs, FCIs are excluded from

negative sentences as well, at least when negation combines with an episodic structure. We will

see in section 4.3 that it is the extensionality of such structures that makes them incompatible

with FCIs.

The anti-episodicity restriction also seems to hold for other languages with lexically

distinct FCIs such as Spanish and Catalan. I illustrate here with data from Quer (1998, 1999):

(6) a *Expulsaron del partido a cualquier disidente. (Spanish; Quer 1999: (21))

expel.3pl from-the party ACC FC dissident

(*‘They expelled any dissident from the party.’)

b * Non expulsaron del partido a cualquier disidente.

(not expel.3pl from-the party ACC FC dissident

(‘They didn’t expel any dissident from the party.’)

(7) a *Li va comprar qualsevol ram. (Catalan; Quer 1998: 220)

her/him aux.3sg to.buy FC bouquet

('*S/he bought him/her any bouquet.')

b *No li va explicar qualsevol conte de por. (Quer 1998: 220)

not him aux.3sg to.tell FC tale of horror

(‘S/he did not tell him any horror tale.’)

The anti-episodicity constraint seems to be at work also in the case of questions with

perfective aspect; an effect found again not only in Greek but in Spanish and Catalan as well.

(8) * Su sistisan opjondhipote thavmasti?

you introduced.perf.3pl FC admirer

(‘Did they introduce any admirer to you?’)

(9) * Et van presentar qualsevol admirador? (Catalan; Quer 1998: 220)

to-you aux.3pl introduced FC admirer

(‘Did they introduce any admirer to you?’)

(10) *Te presentó a cualquier admirador? (Spanish)

you introduced.3pl ACC FC admirer

‘Did they introduce any admirer to you?’

Some parallel data illustrating the same pattern in French are discussed in section 2, where it is

5

shown that it is not negation or questions per se that exclude FCIs but episodicity; if the

episodic structure with negation and in questions is embedded under a habitual or generic

operator (with imperfective aspect) or a modal operator, FCIs become fine.

FCIs typically occur in nonveridical contexts, such as intensional, habitual, generic, and

modal sentences. We see here an occurrence of a FCI in a sentence with the ability modal:

(11) Opjosdhipote fititis bori na lisi afto to provlima.

FC- student can SUBJ solve.3sg this the problem

‘Any student can solve this problem.’

Sentences with FCIs, like (11), have a clear flavor of arbitrariness and indicate that

identity is not so important: whichever student one considers, it doesn’t matter who the student

might be, will have the ability to solve this problem.

The licensing conditions and the interpretation of FCIs have not featured prominently in

the literature (with the notable recent exceptions mentioned above); instead, any has been the

main focus of attention, and inevitably the observations about any will be taken as the point of

departure. Given that any as an API is arguably interpreted inside the scope of negation as in

(12), the question is how any in (11) is interpreted. Since (11) seems to express a generality

about students, it is tempting to consider (13) as a plausible logical representation:

(12) ¬∃x [person (x) ∧ saw (I,x)]

(13) ∀x [student (x) → CΑΝ solve (x, this problem)]

In (13), any is translated as a universal quantifier which scopes over the modal verb; the

sentence in (11) is thus rendered equivalent to “every student can solve this problem”. This is

the view adopted, with varying executions, by a number of scholars, among others, Reichenbach

(1947), Quine (1960), Horn (1972, chapter 3), and recently Dayal (1998), and Sæbø to appear--

the wide scope universalists in Horn’s 1999 terminology. As a consequence of the universalist

analysis, any can no longer be treated as one item, but must be analyzed as quantificationally

ambiguous: under negation, any contributes an existential quantifier, but in free choice

construals any is a universal quantifier.

Two facts appear mysterious in this approach: first, unlike “regular” universals, any

and FCIs are unable to contribute universal quantifiers in isolation; an external operator is

always necessary, e.g. a modal verb as in (11).

(14) a {Every student/all the students/each of the students} solved this problem.

b *Opjosdhipote fititis elise to provlima.

6

*Any student solved this problem.

If any were a true universal, why can’t it contribute ∀ in the absence of modality? The presumed

universal force of any, as well as its very occurrence, seem to be dependent on the availability of

some external operator. If such an operator is absent, as in (14b), so is any’s universality and

grammaticality. Why this is so remains a mystery in the universalist approach.

A universalist, of course, could try to bypass this question by saying that any is a

special kind of universal, which, for some reason, unlike every comes with a requirement that an

external intensional/modal operator be present (see especially Sæbø to appear). While this

correctly describes (one part of) the constraint on the distribution of any, in its FC use, it does

little in the way of advancing our understanding of the source of this constraint, or of explaining

this distribution. And even if we tried to establish the link between the source of the constraint

and the limited distribution by postulating, as in Dayal (1998), inherent modality as the special

feature of any, it is impossible to explain why this inherent modality needs another external

modality for any to be admissible. If any contributed modal alternatives on its own, it should be

able to do so in isolation without requiring the co-occurrence of some other modal operator with

universal force.

Second, the assessment that any, FCIs and universal quantifiers are interchangeable

when construed with can does not seem to be entirely correct. In the sentence below, for

example, they are not:

(15) I epitropi bori na dosi ti thesi se opjondhipote ipopsifio

the committee can subj offer.3sg the position to FC candidate

The committee can offer this job to any candidate. ≠The committee can offer this job to every candidate.

Unlike (11), where can is interpreted as an ability modal, can here is used as an (epistemic or

permissive) possibility modal. The difference seems to affect the interpretation of any and FCIs.

Recall also that any was interpreted existentially in the imperative example in (2). This varying

intepretation is unexpected if any were simply a universal quantifier.

An alternative to the wide-scope-universal idea can be constructed if we assume that any

is a Heimian indefinite. In this view, any doesn’t have quantificational force of its own. It

contributes merely a predicate and a variable to be bound by an intensional Q(uantificational)

operator, binding also a situation or a world variable, and which must be present in the structure

higher up. The semantic representation of (11) in this account would be as in (16):

(16) Q-OPERATOR[w, x] [...any-NP (x,w) ...]

7

If the binding operator has existential force, as permissive and epistemic can above, any will be

interpreted existentially; but if the operator is universal or other (e.g. generic, a Q-adverb like

always, rarely, usually, etc., a universal modal verb, the maximality operator in comparatives)

any will receive the corresponding Q- force. Many authors (the existentialists in Horn’s

terminology) have argued for variants of this idea: Bolinger 1960, 1977, Davison 1980,

Haspelmath 1997, Kadmon and Landman 1993, Lee and Horn 1994, Dayal 1995, Giannakidou

1998, and Tovena and Jayez 1997. A welcome consequence of such an analysis is that, unlike

the wide-scope-universal analysis, it preserves the integrity of any; additionally, it provides a

way to capture the fact that any occurs only in environments where a Q-operator is present, and

that its quantificational force is “parasitic” on that of the Q-operator.

Languages with FCIs lexically distinct from APIs have an important contribution to

make in the any-debate. If it can be shown that in these languages FCIs are indeed universal

quantifiers, then a strong argument for the wide-scope universalist approach can be given. But if

it turns out that FCIs are not universal quantifiers, the wide-scope universalist position and the

ensuing quantificational divide of any will be deprived of support from the quarter where it

would be most likely to be found crosslinguistically. The discussion in this paper should be

seen precisely in this light. We will see that the some version of the second possibility is

realized: FCIs are not universal quantifiers but indefinites.

Of course, FCIs are not only interesting because they are relevant for deciphering the

semantics of a single lexical item in a West Germanic language; they are even more interesting

from the perspective of the general theory of polarity. Perhaps more strikingly than in any other

PI-paradigm, the study of FCIs can make a clear case that the limited distribution is derived

from the lexical semantics of the item in question. This will be the driving force of my proposal.

The basic idea will be that FCIs are intensional indefinites that can be interpreted in a sentence

only if the sentence provides possible worlds which can serve as identity alternatives inducing

variation. Nonveridical sentences (modal, generic, habitual, etc.) are such cases. Veridical and

episodic sentences, on the other hand, do not provide the kind of alternatives needed for the

interpretation of FCIs, hence FCIs are ruled out.

Once evidence is produced that FCIs are not universals, the view of FC-any as a

universal loses much of its appeal, unless independent evidence can be adduced, specific to this

item, which suggests otherwise. As we will see, however, the literature on this question offers no

such evidence -- the most popular diagnostics employed in support of the universalist position

are either flawed, or diagnose plurality and high scalar values rather than universality. In

addition, there are a number of asymmetries between universal quantifiers and FCIs/any which

make assimilating the latter to the former even more problematic.

Before proceeding, let me outline the organization of the discussion. As the title

8

indicates, the main focus of this paper is free choice, and the goal is to provide a concrete

semantics for FCIs. Since most discussion of the alleged universal force associated with free

choice stems from the analysis of any, any will inevitably be included in the discussion as a

secondary focus: partly as a comparison item, and partly again when we consider the

consequences of the proposed analysis and contrast it with the revival of the universalist

position presented in Dayal (1998). We will see in section 5 that it is indeed possible to give an

analysis of any as a unitary item in the approach pursued in this paper.

Moreover, although the discussion of free choice can naturally be embedded into the

more general discussion of indifference and ignorance markers used crosslinguistically with

wh-items, e.g. the -ever paradigm in English whoever, whatever, and also who the hell, offering

a precise semantics of these English expressions is kept beyond the scope of the paper. In fact,

since Greek FCIs are morphologically identical to the -ever paradigm, the obvious next step is

to assume that the core part of the proposed analysis-- i.e. the requirements for nonveridicality

and variation-- will hold for these items too. Although I will not undertake the task of showing

that this is the case, there are a number of recent works (Quer 1998, 1999, 2000, Horn 1999,

2000, von Fintel 2000, den Dikken and Giannakidou 2001) which seem compatible with such a

conclusion.

A final note on the crosslinguistic extensions of the analysis: though recent work on

Spanish and Catalan indicates clearly that the analysis of FCIs proposed here can be directly

extended to account for the relevant paradigms in these languages, I will not attempt a detailed

examination of these facts, since they have been thoroughly discussed in other studies (Bosque

1996, Quer 1998, 1999, 2000). Most importantly, the conclusion in these studies is that FCIs

are indeed indefinites subject to nonveridicality and episodicity constraints (see especially

Quer). Relevant data from Spanish and Catalan will be considered, of course, along the way. I

will further consider data from French: the n’importe qui paradigm can be shown to conform to

the Greek/Spanish/Catalan pattern. More detailed crosslinguistic comparisons must be

eschewed for lack of space, but also because the relevant data are as yet mostly undocumented,

so it is difficult to obtain a clear picture.3 My goal is to provide a semantics for free choice

3 The data reported in Sæbø this issue, for instance, where it is claimed that Scandinavian FCIs differ fromGreek, Spanish and Catalan FCIs in being universal quantifiers, are insufficient to substantiate such a claim. Forone thing, no contrastive examination of FCIs, universals, and indefinites is provided, of the detail pusrued inthis paper, as a motivation for this claim-- rather, Sæbø's position that FCIs are universals is part of hisassumptions. Moreover, the presented data are insufficient to support the claim for the whole language family;although some data from Norwegian are presented, and to a lesser extent from Swedish, there is no discussion ofDanish or Icelandic. The Danish data presented in Vikner (1999) may be taken to suggest that Danish FCIsexhibit a pattern close to the Greek one. Finally, the data provided for Norwegian and Swedish are restricted to alimited class of environments: of the 21 environments given here in section 2.2, Sæbø gives the Norwegianfacts only for five-- modals, conditionals, generics, stative verbs, and comparatives; some data includingpredicative uses of FCIs in Swedish are also included, but constitute a problem for the universalist position,since universals cannot be used as predicates (cf. our discussion in 3.2.1). It is difficult to conduct a

9

precise enough to serve as a reliable testing ground crosslinguistically on a larger scale.

The discussion is organized as follows. In section 2 the distribution of Greek FCIs is

presented. It is shown that these expressions are only grammatical in nonveridical contexts, with

an additional constraint that the nonveridical context be non-episodic (which derives the

distributional differences between FCIs and APIs). In section 3, I review the arguments in favor

of the universal analysis of FCIs and show them to be problematic. Next, a number of

asymmetries are pointed out between universal quantifiers and FCIs which suggest that the two

do not belong to the same semantic class. Finally, arguments are given for an indefinite analysis

of FCIs including their existential readings in many environments (imperatives, conditionals,

modals, free-relatives) and the possibility of donkey-anaphora and predicative uses. In section 4

I propose an intensional semantics for FCIs which successfully derives their quasi-universal

readings and limitation to nonveridical, nonepisodic contexts. In section 5 I explore a possible

extension of this analysis to subtrigging, and consider its overall consequences for any. The

analysis is compared to Dayal’s (1998) recent account, which is shown to be problematic; an

alternative for a uniform indefinite any is proposed instead.

2 FCIs and their licensing conditions

In this section, I examine the licensing conditions of FCIs. It will be shown that FCIs are PIs

proper: they are admitted in a proper subset of the contexts I have characterized elsewhere

(Giannakidou 1998, 1999) as nonveridical, and are excluded from contexts which are veridical.

It will also be shown that FC distribution is further constrained by episodicity. The two

constraints are related in a way that will be made precise in section 4.

2.1 Nonveridicality and polarity

Polarity raises the issue of semantic well-formedness in a model of the grammar like the one in

(17), from Ladusaw (1986: (1)):

(17) grammatical (φ) =def Syn (φ) ∧ Sem (φ),

where Syn is syntactic well-formedness, and Sem is semantic well-formedness

According to (17), grammaticality is a conjuction of syntactic and semantic well-formedness,

and sentences in a language can be ruled out as ungrammatical for semantic as well as syntactic

reasons. In the standard case where syntactic well-formedness entails semantic well-formedness,

i.e. when syntax gives structures that can be assigned a truth value, semantic interpretation

comparison, or make any general claims based on such partial, and to a certain extent conflicting, data.

10

proceeds unobstructedly. But if syntax gives structures that are not semantically well-formed,

i.e. they cannot be assigned a truth value, then these structures will be filtered out by the

semantics.4 The type of ill-formedness that arises with ungrammatical PIs exemplifies precisely

this case. PIs are sensitive expressions, that is, expressions with a lexical semantic ‘deficit’. For

a sentence containing a PI to be assigned a truth value, the lexical semantic requirement of the

PI, relating to this deficit, must be satisfied. If this requirement is not satisfied, the sentence

containing the PI cannnot be interpreted and will therefore be ruled out as ungrammatical (for

more general discussion, see Ladusaw 1996).

In Giannakidou (1997, 1998, 1999) I further argued that the notion of (non)veridicality

is central. PIs are defined as expressions whose distribution is restricted by this property:

(18) DEFINITION 1 (Polarity item).

A linguistic expression is a polarity item iff:

(i) distribution of is limited by sensitivity to some semantic property of the

context of appearance; and

(ii) is (non)veridicality, or a subproperty thereof: β ∈ {veridicality, nonveridicality,

antiveridicality, modality, intensionality, extensionality, episodicity, downward

entailingness}

Simplifying somewhat, the set of properties subsumed under (non)veridicality is presented as a

closed set in (ii) in order to cover the PI-paradigms that have thus far been identified in the

literature. Let it be noted, however, that other properties can be added, e.g. upward entailingness,

anticipating the identification of PIs that might be sensitive to just this property (no such PIs

have yet been identified). Multiple sensitivities are also allowed in this system.

Sensitivity is a form of semantic dependency between a PI and (non)veridicality. This

dependency can be positive or negative. A positive dependency to nonveridicality means that a

PI must be in the scope of a nonveridical operator for the sentence containing the PI to be

grammatical. A negative dependency to veridicality means that a PI must not be in the scope of a

veridical operator in order to be grammatical. The former gives rise to a licensing condition, the

latter to an anti-licensing condition. The abstract format of either case is illustrated below5:

4 I talk here in terms of truth values, but of course something analogous can be said in this model about non-declarative sentences where interpretation is cast in terms of fulfillment conditions (imperatives) or answerhoodconditions (interrogatives).5 I give here very general schemata, without addressing the details of how licensing and anti-licensing translateinto scope conditions. The mapping of licensing onto scope is not as straightforward as one might be inclined tobelieve. For some PIs, for instance, the relevant scope is the local scope of a nonveridical expression , forothers it is the global scope of (with veridical expressions potentially intervening, like in John didn’t say thathe was glad that he saw anybody); for some other PIs licensing corresponds to an escape-the-scope-of condition

11

(19) Licensing by nonveridicality

A polarity item will be grammatical in a sentence S iff is in the scope of a

nonveridical operator in S.

(20) Anti-licensing by veridicality

A polarity item will not be grammatical in a sentence S if is in the scope of a

veridical operator in S.

Licensing expresses a must condition, and it makes a positive prediction as to where the

PI can occur. Anti-licensing, on the other hand, expresses a must not condition and it makes a

predication as to where the PIs cannot occur (see also Ladusaw 1979, Progovac 1994).

Naturally, from knowing that a PI will be ungrammatical in the scope of a veridical operator, we

expect that this PI will be fine in the scope of a nonveridical one, but note that this is a likely

state of affairs rather than a necessary one. Given the kinds of dependencies involved, we expect

empirical differences to arise between items sanctioned through licensing and those sanctioned

through anti-licensing: in section 5.2 we see that the differences between Greek FCIs and any

reflect exactly this difference (see also Giannakidou 1999).

The first mention of veridicality we find in Montague 1969, where it is understood in

terms of existence; see also Dayal 1995 and Lin 1996 for postulating nonexistence as a notion

crucial to PI-licensing. In Giannakidou (1997, 1998, 1999) and Zwarts (1995) veridicality and

nonveridicality are defined in terms of truth as in (21)6:

(21) DEFINITION 2 (Relativized (non)veridicality for propositional operators)

[Giannakidou 1999: 388]

Let c = <cg(c), W(c), M, s, h, w0, f,...> be a context.

i. A propositional operator Op is veridical iff [[ Op p]] c = 1 →[[ p]] = 1 in some epistemic

model ME (x)∈c ; otherwise Op is nonveridical.

ii. A nonveridical operator Op is antiveridical iff

[[ Op p]] c = 1→ [[ p]] = 0 in some epistemic model ME(x)∈c.

(an admittedly more marginal option, for details see Giannakidou 1998, 2000). The type of syntax we need forlicensing (or anti-licensing) will be almost exclusively determined by the specific semantics of the PIs involved.Since FCIs are intensional, as I argue, it is expected that for them licensing indeed translates into a be-in-the-scope-of the licenser condition.6 (Non)veridicality is also defined for determiners but I omit consideration of such cases, as they are not crucialfor the present discussion. Let me also caution that nonveridicality is not identical to intensionality or opacity.There are extensional operators that are nonveridical, for instance, negation which is nonveridical and also non-opaque. On the other hand, as shown in Giannakidou 1999, strong (alethic) and epistemic modalities likebelieve, though intensional, are nevertheless veridical. As indicated in definition 1, intensionality is just asubcase of (non)veridicality.

12

(Non)veridicality is defined as a property of propositional operators in terms of truth

entailment. A propositional operator is veridical iff the truth of Op p in c requires that p be true

in some individual x's epistemic model ME (x) in c. If the truth of Op p in c does not require that

p be true in some such model in c, Op is nonveridical. A nonveridical operator Op is

furthermore antiveridical iff the truth of Op p in c requires that p be false in some epistemic

model ME (x) in c. Antiveridical operators are essentially negative operators and they form a

proper subset of the nonveridical.

Relativization of (non)veridicality to epistemic models is motivated by the need to deal

with the veridicality properties of propositional attitudes (which are treated as propositional

operators, a treatment originating in Hintikka 1962). Epistemic models are construed as sets of

worlds anchored to an individual (the individual anchor in Farkas 1992) representing worlds

compatible with what the individual believes, as in the philosophical tradition where belief and

knowledge states are modeled as sets of possible worlds in terms of accessibility functions

relative to individuals. An epistemic model corresponds to a doxastic function:

(22) DEFINITION 3 (Epistemic model).

(i) A model ME (x) is a set of worlds associated with an individual x, representing worlds

compatible with what x believes.

(ii) ME (s) (w) = {w’: ∀p [s believes p(w) → p (w’)]},

where w is a world of evaluation, and s is the speaker .

In unembedded sentences, of course, the only relevant epistemic agent is the speaker and this is

the only model we consider. But with embedding under propositional attitudes, the model of the

attitude subject is also relevant and can play a decisive role.

Adverbs like yesterday, and the assertion operator in positive sentences are typical

examples of veridical operators; as I stated, the relevant model is the speaker’s but I will not

indicate it here as no other model is relevant.

(23) a Yesterday, Paul saw a snake. → Paul saw a snake.

b ASSERT Paul hit Frank. → Paul hit Frank.

The veridicality of yesterday relates to the fact that it anchors an episodic past event. Logical

conjunction is also veridical, in both arguments positions (see Zwarts 1995 for discussion). PIs

crosslinguistically are ungrammatical in the scope of veridical operators.

The question operator and modal verbs are typical examples of nonveridical operators:

13

(24) a Did Paul see a snake? -/→ Paul saw a snake.

b Paul may have seen a snake. -/→ Paul saw a snake.

PIs are fine in questions and in construction with modal verbs. Negation and without, on the

other hand, are typical antiveridical operators, and PIs are admitted in their scope as well:

(25) a Paul did not see a snake. → It is not the case that Paul saw a snake.

b without Paul seeing a snake. → It is not the case that Paul saw a snake.

Downward entailing operators also form a proper subset of the nonveridical (see

especially Zwarts 1995), but, with the exception of negation and negative quantifiers, the

licensing of PIs in the scope of other downward entailing operators is not that straightforward.7

Moreover, a set of PIs sanctioned exclusively by downward entailingness has not yet been

identified; such a paradigm would have to be grammatical for instance in the scope of

quantifiers meaning few, no and negation, but it should be unacceptable in questions (which are

at most non-monotone; Groenendijk and Stokhof 1984), the protasis of conditionals and the

other intensional and modal environments we see in Table 1 in the next subsection.

Regarding propositional attitudes, the basic observation in Giannakidou 1998, 1999 is

that PIs, any included, are admitted in the scope of directives like want, insist, suggest, etc. but

are excluded from the scope of predicates like believe, dream and say, which I will refer to,

simplifying a bit, as epistemics. The argument was built based on the Greek facts, but data from

other languages support it (see Haspelmath 1997 for a general crosslinguistic picture; also Quer

1998 for Catalan, and Dahl 1999: 673 specifically for the Russian kto-nibud). The contrast

correlates with mood choice (directives select the subjunctive but epistemics require indicative

complements), and seems to be confirmed also by the FCI data. Compare the sentences below,

which are fine, to the ungrammatical sentences in (28):

(26) I Ariadne epemine na afiso opjondhipote na perasi mesa.

7 FCIs, for instance, are not sanctioned in the scope of downward entailing quantifiers in an episodic context:

(i) ??Elaxisti fitites ipan otidhipote.‘Very few students said anything.’

In Giannakidou (1998) occurrences of APIs in the scope of Greek quantifiers meaning few and very few arereported as acceptable, though they are never impeccable. In the approach I formulate here, the ungrammaticalityof (i) follows from episodicity (see section 4.4). Note that the habitual version of (i), given in (ii) with theimperfective lene replacing the perfective ipan, is fine; the habitual allows for PIs:

(ii) Elaxisti fitites lene otidhipote sto mathima. ‘Few students (usually) say anything in class’.

14

the Ariadne insisted.3sg subj let.1sg FC-person subj come.3sg in

‘Ariadne insisted that I allow anyone in.’

(27) I Ariadni mas zitise na tis agorasume {kanena/opjodhipote} vivlio ja ta genethlia tis.

the Ariadne us asked,3sg subj her buy.1pl API/ FCI book for the birthday hers

Ariadne asked us to buy her any (of these books) books for her birthday.

(28) a * O Pavlos pistevi oti akuse {kanenan/opjondhipote} thorivo.

the Paul believe.3sg that .ind heard.3sg API / FCI noise

* Paul believes that he heard any noise.

a' (Paul said that he heard a noise.)

b * Onireftika oti irthe {kanenas/opjosdhipote} idravlikos.

dreamt.1sg that .ind came.3sg API / FCI plumber

* I dreamt that any plumber came in.

b' (I dreamt that a plumber came in.)

c *I Ariadne ipe oti akuse {kanenan/opjondhipote} thorivo.

the Ariadne said.3sg that .indicative heard.3sg API/ FCI noise

* Ariadne said that she heard any noise.

c' (Ariadne said that she heard a noise.)

This contrast provides one of the strongest arguments for taking nonveridicality to be

the notion regulating the distribution of PIs. The more narrow alternatives of downward

entailingness and negation are silent on these facts.

Any, as we see, is also inadmissible in the scope of epistemic attitudes, but is fine with

directives. The following examples make the contrast with directives even more clear (told in

(30b) is used as equivalent to suggest , or ask in this context):

(29) a John would like us to buy any book on this list for his birthday.

b John told us to buy any book on this list for his birthday.

c * John believes that we bought any book on this list for his birthday.

d * John dreamt that we bought any book on this list for his birthday.

Likewise, Quer (1998, 1999) confirms the contrast for Spanish and Catalan FCIs. The mood

contrast is also confirmed, to illustrate (see also Quer 1998:179):

(30) a * La Paola creu que desapareix qualsevol llengua minoritària. (Catalan)

b *Paola cree que desaparece cualquier lengua minoritaria. (Spanish)

Paola believe.3sg that disappear.ind.3sg FC language minority

*Paola believes that any minority language is disappearing.

15

(31) a La Paola vol evitar que disaparegui qualsevol llengua minoritària. (Catalan)

b Paola quiere evitar que desaparezca cualquier lengua minoritaria. (Spanish)

Paola want.3sg avoid disappear.subj.3sg FC language minority

Paola wants to prevent any minority language from disappearing.

Epistemic attitudes are veridical according to definition 2: the truth entailment holds in

the believer’s, dreamer’s or sayer’s epistemic model ME (subject). Directive attitudes, on the

other hand, do not give rise to a truth implication with respect to either the speaker or the

embedded subject, and are thus nonveridical. I will not give further details here, but see

Giannakidou (1998, 1999) for extensive discussion. The crucial fact is that the licensing of

FCIs in propositional attitudes follows the pattern observed for other PIs; any appears no

different in this respect.

The fact that veridical attitudes are incompatible with FCIs has an important

consequence regarding the role of intensionality in licensing FCIs. Approaches that take this

notion to be central (Sæbø this issue, and less explicitly so Dayal 1998) would expect FCIs to

be grammatical with both epistemic and directive attitudes since in both cases we are dealing

with intensional domains. The contrast just observed indicates that the crucial factor is

nonveridicality rather than intensionality.

2.2 The distribution of FCIs: nonveridicallity and episodicity

In this section I illustrate the distribution of FCIs. FCIs are generally fine in nonveridical

contexts, but are excluded from veridical ones. Episodicity is shown to further constrain the set

of nonveridical enviroments that allow for FCIs. This observation, as we noted at the outset,

holds for Greek as well as Spanish, Catalan, and as we shall see here, French.

An overview of the distribution is given in Table 1. The table summarizes the Greek

data contrastively with APIs and any. All PI paradigms are excluded from veridical contexts (the

five bottom rows). But the distribution of FCIs, though overlapping with that of any and APIs

for the most part, is not identical to these. The differences between APIs and FCIs are discussed

elsewhere (Giannakidou 1997b, 1998) and since they are not directly relevant for our purposes

here, I omit consideration. The differences with any will receive an explanation in section 5.2.

Below, some examples are given as an illustration of successful occurrences of FCIs:

16

Table 1 Contrastive distribution of any, FCIs, and APIs

Environments any FCIs APIs

1. Episodic negation OK * OK2. Episodic questions OK * OK3. Conditionals OK OK OK4. Restriction of universal OK OK OK5. Future particle/will OK OK OK6. Modal verbs OK OK OK7. Directive intensional verbs % OK OK8. Imperatives OK OK OK9. Habituals OK OK OK10. Disjunctions * OK OK

11. isos/perhaps * OK OK12. Stative verbs OK OK *13. Generics OK OK *14. NP-comparatives OK OK *15. monon/only OK * *16. Negative factives OK * *17. Affirmative episodic sentences * * *18. Existential constructions * * *19. Epistemic intensional verbs * * *20. Progressives * * *

21. Factive verbs * * *

Protasis of conditionals

(32) An kimithis me opjondhipote, tha se skotoso.

if sleep.2sg with FC-person fut you kill.1sg

If you sleep with anybody, I’ll kill you.

Directive intensional verbs (selecting subjunctive )

(33) I Ariadne epemine na afiso opjondhipote na perasi mesa.

the Ariadne insisted.3sg subj let.1sg FC-person subj come.3sg in

‘Ariadne insisted that I allow anyone in.’

Modal verbs

(34) Bori na bike opjosdhipote mesa.

can.3sg subj entered.3sg FC-person in

‘Anyone may have come in.’

(35) Boris na dhanistis opjodhipote vivlio.

can. 2sg subj borrow.2sg FC book

‘You may borrow any book.’

17

(36) Opjosdhipote fititis bori na lisi afto to provlima.

FC student can.3sg subj solve.3sg this the problem

‘Any student can solve this problem.’

Implicit permission modals

(37) Opjosdhipote ine kalodhexumenos sti sinantisi apopse.

FC-person be.3sg welcome in-the meeting tonight

‘Anyone is welcome at the meeting tonight.’

Imperatives

(38) Pare opjodhipote milo.

take.2sg FC apple

‘Take any apple.’

Future

(39) I Maria tha paralavi opjondhipote proskleklimeno omiliti apo to aerodromio.

the Maria will pick.3sg FC invited speaker from the airport

Maria will pick up any invited speaker from the airport.

Habituals

(40) Sinithos dhiavaze opjodhipote vivlio me megali prosoxi.

usually read.3sg FC book with great attention

‘S/he usually read any book very carefully.’

Comparatives (phrasal and clausal)

(41) a I Ana trexi grigorotera apo opjondhipote stin taksi tis.

the Ann run.3sg faster than FC-person in-the class hers

‘Ann runs faster than anybody in her class.’

b I Ana egrapse kalitera apoti perimene opjosdhipote apo mas.

the Ann wrote.3sg better than expcted.3sg FC-person from us

‘Ann did better (in the exam) than any of us had expected. ’

Generic sentences

(42) Opjadhipote ghata kinigai pondikia.

FC cat hunt.3sg mice

‘Any cat hunts mice.’

18

FCIs are ungrammatical in veridical sentences, e.g. positive episodic sentences like (43)

and existential sentences. Though I give examples with singular FCIs, the ungrammatical data

presented here can be reproduced with plural FCIs; I stick to the singular for simplicity. (Recall

the data from section 1 indicating that Spanish and Catalan FCIs behave on a par). French FCIs

exhibit the same pattern, as we see below:

Affirmative episodic

(43) * Xthes to apojevma, idha opjondhipote ston kipo.

yesterday the afetrnoon saw.perf.1sg FC-person in-the garden

(*Yesterday afternoon I saw anybody in the yard)

(44) *Marie a vu n’importe qui cet après-midi. (French)

Mary has seen FC-person this afternoon

(*Mary saw anybody this afternoon).

Existential sentences

(45) *Ipirxe opjodhipote vivlio pano sto trapezi.

(*There was any book on the table.)

(46) * Il y a n’importe quel livre sur la table. (French)

it was FC- book on the table

(*There was any book on the table.)

It should be noted, then, that veridicality rules out FCIs and any from existential constructions,

and it is not necessary to invoke contextual vagueness (as in Dayal 1998:463). If the existential

construction contains an operator of the appropriate kind, e.g. a modal, FCIs can be improved,

as indicated in the following example from Catalan; note also the improvement with any:

(47) Aquí podría haber cualquiera de tus colegas. (Spanish; Quer 1999: (6))

here could to-have FC of your colleagues

There could be any of your colleagues there.

Admittedly, existential constructions, even with modal verbs, are still resistent environments for

FCIs because it can be assumed that existence scopes over modality, hence allowing the

veridical inference to survive. Although the example above is fine for English, it is not always

the case that by inserting a modal operator we have improvement of any-- e.g. *There could be

any books on the table is still bad, and one can think of many similar examples. The point to

note here is that, to the extent that any/FCIs are admissible in existential constructions with

modals, they contrast with true universals which are consistently ungrammatical in parallel

19

structures, regardless of whether we have a modal or not:

(48) * Aquí podría haber todos de tus colegas. (Spanish)

here could to-have all of your collegues

* There could be {all/every one of} your collegues there.

* There are {all/every one of} your collegues there.

So we are witnessing a relatively clear contrast between universal quantifiers and FCIs; more

asymmetries between universal quantifiers and FCIs are discussed in section 3.2.

FCIs are also excluded from only and factive domains, just like the items of the API

paradigm. Again, the crucial factor is that factives and only give rise to veridical entailments:

(49) * Monon i Theodora idhe {opjondhipote/kanenan} fititi.

only the Thedora saw.3sg FC/ API student

‘Only Theodora saw any students.’

(50) a * Ekplisome pu exi {opjondhipote/kanenan} filo.

be-surprised.1sg that has FC / API friend

‘I’m surprised she has any friends.’

b *Xerome pu exi {opjondhipote/kanenan} filo.

be-glad.1sg that has FC / API friend

‘*I’m glad she has any friends.’

FCIs thus differ from any which may occasionally be accepted with only, and in the

complement of a negative factive, as we see. Grammaticality in these cases can arguably be

accounted for by invoking a weaker condition based on negative implicature (Linebarger 1980;

cf. what I referred to as “indirect licensing” in Giannakidou 1999). Dayal (1998: 442) presents

any in negative factives as a criticism against nonveridicality, but it should be clear from the

above examples that nonveridicality is indeed respected by FCIs and APIs besides any. Given

that comparable PIs in other languages are also uniformly excluded from factives (e.g. Catalan,

Quer 1998, and Haspelmath 1997), it seems fair to say that any exemplifies the marginal case

rather than the typical one.8

8 Von Fintel 1999 proposes a semantic alternative of why any is admitted in the scope of negative factives byanalyzing negative factives as downward entailing (and thus positive factives as upward entailing). A problemwith this approach is that the notion of entailment employed for factives is non-standard in being dependent onsatisfying certain presuppositions, and at any rate different from classical entailment, which used to derivemonotonicity patterns. But even if downward entailment could somehow be made to be part of the meaning of anegative factive, the problem remains that crosslinguistically the scope of a factive (negative or positive) isdisfavored as a PI-environment. If downward entailment is relevant at all for PI-licensing, the marginality of PIs

20

In addition, as we see in (51)-(54), FCIs are not grammatical with progressives (though

they are fine with habituals as in (40); see also Vikner 1999: 34-35 for a parallel with Danish

PIs). This is expected under the analysis of the progressive as veridical in Giannakidou and

Zwarts (1999). Note that any is also ungrammatical:

Progressives

(51) *Egrafa olo to proi opjodhipote grama.

wrote.imperf.1sg all the morning FC letter

‘*I was writing any letter all morning.’

(52) * El David està escoltant qualsevol disc. (Catalan; Quer 1998:221)

the David be.3sg listening-to FC record

*David is listening to any record.

(53) Está hablando con cualquiera de sus empleados. (Spanish;Quer 1999: (22))

be.3sg speaking with FC of his employees

*S/he is talking to any of his employees.

Crucially, FCIs are ungrammatical in certain nonveridical contexts, namely negation and

interrogatives, when these are episodic. Any is apparently not sensitive to this distinction.

Spanish and Catalan FCIs behave on a par with Greek (recall the data in section 1).

(54) * I Roxani dhen idhe otidhipote.

the Roxanne not saw.3sg FC-thing

‘Roxanne didn’t see anything.’

(55) * Aghorases opjodhipote vivlio?

bought.2sg FC book

Did you buy any book?

In French, although the FCI is not ungrammatical, it is accepted only with a special reading

equivalent to English ‘just anybody’. This reading has been characterized as 'indiscriminative' in

Horn (2000).

with negative factives is not expected. Rather, the blocking effect of factivity is expected under the assumptionthat it yields veridicality. The negative implicature account, then, as formulated in Giannakidou 1999, seems thepreferable option because it is most compatible with the marginality of PI-occurrences with negative factives.

21

(56) Marie n'a pas parlé a n' importe qui cet après midi. (French)

Mary not has talked to FC person this afternoon

Mary didn’t talk to just anybody this afternoon.

This sentence does not mean that Mary didn’t talk to anybody, but the opposite: it is stated that

Mary did talk to somebody, and additionally, it is implicated that the person she talked to was

not just some random person. Similar uses of FCIs are observed in the other languages under

consideration, to be discussed later on. In these uses, which typically occur with negation, any

and the FCIs have an interpretation like “an arbitrary individual” which seems like a natural

extension of the FC meaning “any individual, it doesn’t matter which one”. Indiscriminative

uses are marked in English with just, and in Greek with the indefinite article as we note in 3.2,

but this marking is not obligatory; French, as we see, does not mark this special use in a

particular way.

Likewise, n’importe qui is ungrammatical in episodic questions:

(57) * Est-ce que Marie a parlé avec n'importe qui cet après midi?

is it that Mary has talked with FC-person this afternoon

Did Mary talk to anybody this afternoon?

Hence the French FCI n'importe qui mostly patterns with its Greek, Catalan and Spanish

counterparts regarding episodicity. 9

It is not a general incompatibility between FCIs and negation or questions that rules out

FCIs in the sentences above; rather, it is episodicity that must be held accountable. Notice that in

the following examples with the imperfective, FCIs are fine (similar results obtain in

9 A reviewer presents the grammaticality of qui que ce soit under negation as a case of a FCI which does notconform to the episodicity generalization:

(i) Marie n' a pas parlé a qui que ce soit cet après midi.Mary didn’t talk to anybody this afternoon.

However, it is not obvious that qui que ce soit is a FCI. Rather, qui que ce soit appears in traditional negativepolarity items environments, which suggests a characterization as 'NPI' (or more generally 'API') rather than FCI(see also Haspelmath 1997). Cases like (i) therefore have no bearing on the anti-episodicity constraint for FCIs.

Qui que ce soit may also appear as a relative clause modifier in non-negative environment, as in (ii):

(ii) Qui que ce soit qui t'ait dit ça, il a eu tort.who that it be that to-you said that, he was wrong

{Whoever/Anyone who} said that to you was wrong.

This is an instance of subtrigging, to be discussed in section 5.1. If qui que ce soit is indeed possible in suchconstruals, then, at most, it resembles any which has both API and FCI uses. Just like any, then, it shows nosensitivity to episodicity. I am most thankful to Danièle Godard for her assistance with the French data.

22

Spanish,Catalan and French).

(58) Tin epoxi ekini, miluse i Ariadne me opjondipote fititi?

the time that talked.impf.3sg the Ariadne with FC student

‘At that time, did Ariadne use to talk to any students?’

(59) Sinithos dhen milai me opjondipote fititi.

usually not talked.impf.3sg with FC student

‘Usually, she doesn’t talk to any students.’

The following, then, seem secure conclusions:

(a) FCIs are polarity items (PIs) in the sense of Giannakidou 1998, 1999: they are

admitted in nonveridical contexts but are excluded from veridical ones.

(b) Conclusion (a) holds for any too, with the exception of only and negative factives

which allow any despite being veridical, due to the option of indirect licensing by a negative

implicature that is available in these cases (but not in Greek).

(c) In addition to veridicality, FCIs appear to be allergic to episodicity: they are not

acceptable in episodic sentences, but they become licit if we insert an operator inducing plurality

of events or situations. Unlike FCIs, any is not sensitive to episodicity.

Based on these conclusions, we can posit (60) as the distributional constraint on FCIs:

(60) Licensing condition on FCIs

A FCI α is grammatical in a sentence S iff:

(i) α is in the scope of a nonveridical operator β; and

(ii) S is not episodic.

As I said at the outset, postulating a licensing condition is only doing half the job, and,

unless we address the issue of why it is that the PI is subject to the condition it is, a licensing

condition is merely a filter stipulating that a PI α must occur in some context β as a derivation

external global constraint. We don't want PIs to be “special” items in this way. So we have to

address the sensitivity question (see especially Israel 1996, Giannakidou 1998 and Tovena

1999): why are FCIs subject to the conditions in (60)? I will be concerned with this question in

the remainder of the paper and will try to derive (60) from the semantics of free choice.

3 Free choice is not universal quantification

The first step in understanding the lexical semantics of FCIs is to settle the issue of their

quantificational force. In this section I consider potential evidence in favor of the universal

analysis and show it to be unconvincing. Instead, a number of facts are presented in sections

23

3.2. and 3.3. which can be explained only if we assume that FCIs are indefinites.

3.1 Two apparent arguments that FC may be universal quantification

Let us consider first two cases which suggest that FCIs may be universal quantifiers. These

have been extensively discussed in the literature on any; for recent discussion see Dayal (1998).

3.1.1 Almost/absolutely modification

Universal quantifiers, but not existentials like some, admit modifiers like almost and absolutely

(for the original observation see Dahl 1970 and Horn 1972). The examples below show that

FCIs can indeed be modified by sxedhon ‘almost’ on a par with universals:

(61) Sxedhon opjosdhipote fititis bori na lisi afto to provlima.

almost FC student can subj solve.3sg this the problem

‘Almost any student can solve this problem.’

(62) a Irthe sxedhon kathe fititis.

came.3sg almost every student

Almost every student came.

b * Irthe sxedhon kapjos fititis.

came.3sg almost some student

* Almost some student came.

It has been pointed out numerous times in the literature, however, that the results of this test do

not necessarily guarantee universality (see among others McCawley 1981, Horn and Lee

(1995), Horn (1999), and Giannakidou 2000). For one thing, almost can modify predicates:

(63) Ine sxedhon ilithios.

is.3sg almost idiot

‘He is almost an idiot.’

Furthermore, almost can modify high value cardinals like two thousand students, or even low-

value ones, as long as they can be contextually interpreted as denoting high values:

(64) Almost two thousand students participated at the demonstration.

(65) A local phone call from the hotel room costs almost one dollar!

The sentence in (65) has the neutral mathematical interpretation of ‘rounding up’ the amount of

money to one dollar, and additionally, it licenses an inference (probably an implicature) that one

dollar is a lot to pay for a local phone call. This inference suggests that almost associates with

24

high values (Horn and Lee 1995), and it further indicates that even low values can be interpreted

as high if the context allows it. (This, of course, is expected because scales are pragmatic

constructs.) Likewise, when almost modifies a predicate, e.g. (63), it associates with a value

close to the supremum of the predicate’s lattice structure.

Hence, sentences like the above suggest that the prerequisite for almost modification is

some kind of implicated plurality rather than universality. This conclusion of course does not

constitute an account of how almost modification exactly works, but providing such an account

is not a task to be undertaken in this paper. For our purposes, it merely suffices to see that the

fact that FCIs are modifiable by almost is not necessarily indicative of universal force, but

rather, it may be taken to support the observation that FCIs are somehow pluralized. 10

As further evidence against the validity of the almost/absolutely test as a diagnostic for

universality, consider that, despite the parallelism between FCIs and universals in sentences like

(61), it is not always the case that almost/absolutely modify FCIs. We see below that the Greek

apolitos ‘absolutely’ consistently fails to modify FCIs:

(66) a Absolutely every student student can solve this problem.

b * Apolitos opjosdipote fititis bori na lisi afto to provlima.

? Absolutely any student can solve this problem.

There are also instances where almost modification fails, as indicated in the examples below

from Spanish (Quer 1999: (7b)) and Greek:

(67) *Si lees casi cualquier periódico extranjero, estás mejor informado.

if read.2SG almost FC newspaper foreign be.2SG better informed

‘?? If you read almost any foreign newspaper, you are better informed.’

(68) *An dhiavasis sxedhon opjadhipote kseni efimeridha, tha ise kalitera enimeromenos.

?? If you read almost any foreign newspaper, you are better informed.’

(Since almost modification is pragmatically regulated, it would seem more appropriate to mark

the modification failures with '#" rather than "*', but I will not insist on this here.) In

10 In a way that regular indefinites are not, since they are consistently ungrammatical with almost:

(i) *Almost students know the answer.(No Greek equivalent because bare plurals cannot be used generically in subject position.]

(ii) *Sxedhon enas fititis bori na lisi afto to provlima.almost a student can subj solve.3sg this the problem*Almost a student can solve this problem.

The plurality of FCIs, absent with regular indefinites, has to do with the fact that they make us consideralternatives and is captured in the notion of identity alternatives in section 4.

25

conditionals, both existential and universal readings arise with FCIs and any, as we note in

section 3.3.2. One can say, then, that almost modification fails in the above examples because

the FCIs are interpreted existentially, which in itself argues against the assimilation of FCIs with

true universals. In support of this, witness that almost modification fails consistently in other

contexts licensing existential readings, e.g. imperatives:

(69) ?? Patiste sxedhon opjodhipote pliktro ja na sinexisete.

press.2pl almost FC key for subj continue.2pl

?? Press almost any key to continue.

Consider, finally, examples like the following, from Horn (2000: (42) and (44)):

(70) a If you eat absolutely any meat, you are not a vegeterian. (API-any)

b If you eat absolutely any meat, you are not a kosher Jew. (FC-any)

(71) a If you go to bed with almost anyone, you should use a condom. (API-any)

b If you go to bed with almost anyone, you better use two. (FC-any).

In these examples we see that both existential (API-any) and universal (FCI-any) readings are

modifiable by almost and absolutely. If this is so, then it is hard to see what almost/absolutely

modification actually diagnoses, and almost impossible to maintain that it is universality.

3.1.2 Exceptive phrases

Exceptive phrases are known to only be associated only with universal quantifiers (see von

Fintel 1994 and references therein). FC-any is claimed to be universal because it can be

associated successfully with an exceptive phrase (Dayal 1998). FCIs behave on a par:

(72) To provlima ine toso efkolo, oste opjosdhipote fititis bori na to lisi, ektos-apo to Jani.

the problem is so easy that FC student can subj it solve.3sg but the John

‘The problem is so easy that any student but John can solve it.'

(73) Milisa me kathe fititi ektos-apo to Jani.

talked.1sg with every student but the John

I talked to every student but John.

Singular existentials, as we see below, cannot be construed with ektos-apo ‘but/except’:

(74) * Milisa me kapjon fititi ektos-apo to Jani.

talked.1sg with some student but the John

I talked to some student {but/except} John.

26

Von Fintel (1994) proposes a semantics for but that derives the distributional restriction on

universal associates (von Fintel 1994: 108) in the following way:

(75) Det A [[ but ]] R P ⇔ D (A-R) (P) ∧ ∀S (D (A-S) (P) → R⊆ S)

The above is intented to capture the two main features of but: its domain subtraction (the first

conjunct), and a uniqueness condition ensuring that there will be a unique exception set (the

second conjunct). The argument goes that since only universal determiners guarantee the

existence of a unique exception set, we expect that only these will be compatible with but.

Yet the generalization that but and exceptive phrases associate with universal determiners

does not seem entirely secure. For one thing, there are universal determiners that are

incompatible with but. Cases in point are both, neither, and each:

(76) a *Bill talked to both students but John.

b *Bill talked to neither student but John.

c *Bill talked to each student but John.

Likewise, plural definites do not admit but-phrases despite the fact that they have unique

exception sets, just like universal quantifiers:

(77) * The students but John and Bill attended the meeting.

On the other hand, Hoeksema (1990) and Horn (1999, 2000) report cases of but modifying

non-universal quantifiers like the mass little:

(78) a We had little choice but to comply.

b Landowners could do little but accept their fate. (Horn 1999: (23))

c We are achieving little but increased and forced evacuation of the ethnic

Albanians. (Horn 1999: (23))

Horn (1999, 2000) also presents cases of but phrases modifying a wh-word.

(79) Who but Tipper could be excited by Al?

In this case, an implicature is licensed that Al is so boring that nobody other than Tipper, his

wife, would be excited by him. Similar examples can be reproduced for Greek:

27

(80) Pjos ektos-apo to Jani tha boruse na pi kati tetjo?

who but the John fut could subj say.3sg something such

‘Who but John could say anything like this?’

Again, we have the implicature that nobody but John could say something like this. On the basis

of such examples, Horn hypothesizes that a proposition with but is felicitous only if its

nonexceptive counterpart is conventionally used to express a universal proposition (Horn 1999).

If this hypothesis is correct, then but can no longer be used as a diagnostic of asserted universal

force; this is fully compatible with the analysis of free choice I will propose here, since

universality comes in the requirement that we exhaust all i-alternatives, and this requirement is a

presupposition rather than an assertion (section 4.1).

More deviances from the generalization that exceptive phrases only associate with

universals are encountered when we move across languages. The Greek ektos-apo, for instance,

is quite successful with non-universal quantifiers meaning most, almost every:

(81) a Milisa me tus perissoterus fitites ektos-apo to Jani.

‘? I talked to most (of the) students {but/except} John.’

b Milisa me olus sxedhon tous fitites ektos-apo to Jani.

‘I talked to almost all the students {but/except} John.’

As we see, ektos apo is glossed as but/except because the lexical distinction does not exist in

Greek, and this perhaps is to be held accountable for the judgments above. Note, in this

connection, that there are also empirical differences between but and except in English. I will

gloss over these differences because they are not crucial in making the general point I want to

make here. It is also not crucial to propose an explanation of how exactly exemptive phrases

work, but assume that some version of Horn’ s hypothesis is correct. Once again, then, the

compatibility of FCIs with exceptive phrases suggests that we are dealing with a plural value, yet

not necessarily at the level of assertion.

We may conclude, then, that although we have confirmed our original observation that

FCIs have pluralized interpretations, the facts presented here do not actually provide evidence

that FCIs are universal quantifiers. To this conclusion, let me add one more argument.

FCIs crosslinguistically involve a wh-paradigm; yet it is not immediately obvious what

the connection may be between wh-words and universal quantifiers. In fact, many scholars

argue against such a connection (see especially Jacobson 1995 for arguments against the

universal analysis of free relative wh-words, which is the paradigm that Greek and other

languages (cf. English -ever free relatives) specifically employ for FC-formation). On the other

hand, there have been numerous attempts to link wh-words to existential quantifiers and

28

indefinites (Karttunen 1977, Berman 1991). If FCIs are indeed universals, then it seems quite

peculiar, and certainly not compositional, that natural languages systematically employ a

paradigm for FCIs that is linked to indefiniteness instead of universality. But if FCIs are

indefinites, this strategy seems more expected.

Next, I discuss some asymmetries between universal quantifiers and FCIs which allow

the stronger conclusion that FCIs are not universals but indefinites.

3.2 Further asymmetries between univesal quantifiers and FCIs

We have already noted in 2.2. an asymmetry between universal quantifiers and FCIs regarding

existential constructions: FCIs but not universals occur happily in these constructions, provided

that a nonveridical operator is present. Three additional asymmetries will be presented here: (a)

unlike universal quantifiers, FCIs can be used as predicate nominals; (b) FCIs but not universal

quantifiers can bind pronouns in donkey-anaphora; (c) FCIs, unlike universal quantifiers,

cannot take inverse scope over quantifiers that c-command them at s-structure.

3.2.1 Use as predicate nominals

It is well known that universal quantifiers with every cannot be used as predicate nominals:

(82) Frank is {a/*every} friend of mine.

Partee 1987 discusses restrictions on the availability of type-shifting to predicative (type <e,t>)

interpretations. She argues that type lowering from the generalized quantifier (GQ) type

<<e,t>,t> to <e,t> is not allowed in the case of universals. The reason has to do with the workings

of the relevant lowering operation BE:

(83) Type lowering (Partee 1987)

[[ BE (GQ) ]] = {x| {x} ∈ [[ GQ ]] }

BE picks out all the singletons from a GQ and collects their elements in a set. When we apply

BE to the GQ denoted by a singular indefinite like a student, the result is fine because the

extension of student consists of singletons containing one student. But when we apply BE to

the GQ denoted by every NP the result will be the empty set in case the extension of the NP

consists of two or more individuals, since in this case the GQ contains no singletons.

Unlike with every, BE seems to work with FCIs, which can indeed be used in predicative

positions. This indicates that the GQ denoted by a FCI contains singleton sets. Most

interestingly, in such uses FCIs must be construed with the indefinite article enas ‘a/one’ (the

kathe opjodhipote ‘every any’ string is unattested):

29

(84) Dhen ine enas opjosdhipote daskalos. (Ine o kaliteros!)

not be.3sg a FC teacher is the best

‘He is not just any teacher. He is the best teacher!’

(85) Dhen ime enas opjoshipote ego ja na mou ferese etsi! (Ime o aderfos su!)

not be.1sg a FC I for subj me treat.2sg so am the brother yours

‘I am not just anybody to be treated this way. I am your brother!’

Such predicative uses are also possible with any, as we see. Quer 1999 reports a similar

observation about Spanish FCIs: they can be used predicatively, and can only be construed with

the indefinite determiner (Quer 1999: (4)):

(86) No está {una/*la/*toda/*cada} revista cualquiera.

not is a the all each magazine FC

'This is not just any magazine.'

The predicate uses we are considering here are indiscriminative uses; as we saw in 2.2., such

uses are observed with FCIs crosslinguistically (cf. the French data in (56)). Indiscriminative

uses are not necessarily predicative, e.g. Don't buy just any book, but a good one.

If FCIs were universals, we wouldn’t expect to find them as predicates; we also

wouldn’t expect to find them with the indefinite determiner. The behavior of FCIs regarding

these facts makes it natural to pair them with indefinites rather than universals.

3.2.2 Donkey anaphora

Universal quantifiers are static and do not license donkey anaphora, but existentials are

dynamic: they can bind variables outside their scope. The difference is illustrated below:

(87) * I fitites pu aghorasan kathe vivlio1, na mu to1 diksun amesos.

the students that bought.3pl every book, subj me it show.3pl immediately

(* The students that bought every book1 should show it1 to me immediately.)

(88) I fitites pu aghorasan ena vivlio1, na mu to1 diksun amesos.

the students that bought.3pl a book, subj me it show.3pl immediately

The students that bought a book1 should show it1 to me immediately.

FCIs, and similarly any, pattern with existentials rather than with universals; Quer 1999, 2000

presents similar examples from Spanish and Catalan which support this point (Quer 1999: (5)):

30

(89) I fitites pu aghorasan opjodhipote vivlio1, na mu to1 diksun amesos.

the students that bought.3pl FC book, subj me it show.3pl immediately

'The students that bought any book1 should show it1 to me immediately.'

(90) Si llama cualquier cliente2, le2 diré que no estás. (Spanish)

if call.3sg FC client him tell fut.1sg that not be.2sg

‘If any client calls, I will tell him that you are not here.'

In the same spirit, Lee and Horn (1994) present data showing that any can be unselectively

bound by an operator just like regular indefinites, and unlike universals:

(91) a If a farmer1 owns a donkey, he1 beats it.

b If any farmer1 owns a donkey, he1 beats it.

c *If every farmer1 owns a donkey, he1 beats it.

If FCIs and any were universal quantifiers we would not expect them to exhibit these

distinctive features of indefinites.11

Note that the nonveridicality requirement is obeyed in the examples above, because the

restrictions of plural definites (and every for that matter) are nonveridical (Giannakidou

1999:396-404). Hence it is not known in the context that students that bought any books exist:

The students that bought any book should show it to me immediately, but I doubt whether there

will be any such students is a non-contradictory sentence.

3.2.3 The possibility of taking inverse scope over the licensing operator

Universal quantifiers can generally scope over c-commanding quantifiers or other operators in

the same clause, a possibility known as inverse scope. Sentence (92) has the two readings in

(93);

(93b) represents the inverse scope reading, where students vary with professors:

11 A reviewer notes that when any is modified by almost or an exceptive phrase, donkey anaphora is blocked:

(i) a ??If Sue can buy almost any book, she will buy it.b ??If Sue can buy any book except War and Peace, she will buy it.

Though I will not provide an explanation, note that this fact cannot necessarily be taken as an indication foruniversality; we see below that singular indefinites exhibit the same effect:

(ii) ??If Sue can buy a book, but not War and Peace, she will buy it.

31

(92) Kapjos fititis tha paralavi kathe proskeklimeno omiliti apo to aerodromio.

some student fut receive.3sg every invited speaker from the airport

Some student will pick up every invited speaker from the airport.

(93) a ∃x [student(x)∧ FUT ∀y [invited-speaker(y) → pick-up(x,y)]]

b ∀y [invited-speaker(y) → FUT ∃x [student(x) ∧ pick-up(x,y)]]

To derive the surface reading, I assume that the highest quantifier has been QRed to an IP-

adjoined position at LF; the second quantifier is adjoined to VP, thus lower than the future tense

operator, as in (94) (for recent arguments as to why we need VP as a possible adjunction site for

polarity items see Merchant 2000). In the inverse scope reading, kathe proskeklimeno omiliti is

QRed and adjoined to IP, and the subject kapjos fititis (95) is also adjoined to IP in a lower

position. (The verb moves to I0 in both cases).

(94) surface scope

[IP kapjos fititis1 [IP tha paralavi [VP kathe proskeklimeno omiliti2 [VP t1 tv t2 ...]]]]

(95) inverse scope

[IP kathe proskeklimeno omiliti2 [IP kapjos fititis1 [IP tha paralavi [VP t1 tv t2 ...]]]]

There are certain contexts where in fact only the inverse scope reading makes sense.

One such case is illustrated with the sentence below:

(96) Ja tin pirosvestiki askisi stis 4 i ora,

for the fire exercise at 4 o'clock

tha topothetisume enan paratiriti brosta se kathe eksodo.

fut put.1pl an observer in-front-of every exit

'For the fire drill at 4 o'clock, we will station an observer in front of every exit.

(97) a # ∃x [observer (x) ∧ FUT ∀y [exit (y) → station (we, x, y)]]

b ∀y [exit (y) → FUT ∃x [observer (x) ∧ station (we, x, y)]]

On the surface reading, there is one particular observer who will be stationed in front of every

single exit, which is absurd given that one individual cannot be in more than one place at a time.

On the inverse reading, the sentence says that a different observer will be put in each exit. This

is the only reasonable reading, and indeed the only reading (96) has.

Consider now what happens if we replace the every-argument with a FC-argument.

32

(98) Kapjos fititis tha paralavi opjondhipote proskeklimeno omiliti apo to aerodromio.

Some student will pick up any invited speaker from the airport.

Unlike (92), the sentence above does not have the wide scope universal reading in

(93b). Rather, it has an interpretation in which some student picks up a invited speaker in all

situations that we may consider. This reading is not necessarily the one in

(93a), but it can be taken to correspond to (99) below, where the FCI is represented as an

indefinite inside the scope of the future operator and bound by it. The future contributes a

universal quantifier over possible worlds, and the FCI ends up being interpreted universally.

Kapjos 'some student', on the other hand, is interpreted outside the scope of the future as

referring to a particular student.12

(99) ∃x [student (x) ∀w,y [ invited-speaker(y, w) → pick-up(x,y,w)]]

Consider now an example parallel to (96).

(100) Ja tin pirosvestiki askisi stis 4 i ora,

for the fire exercise at 4 o'clock

tha topothetisume enan paratiriti brosta se opjadhipote eksodo.

fut put.1pl an observer in-front-of FC exit

'? For the fire drill at 4 o'clock, we will station an observer in front of any exit.

Again, a wide scope universal reading of the FCI and any is missing, and the FCI can be taken

to have a bound reading. In this case, enan paratiriti 'an observer' is also bound by the future:

(101) ∀w,x,y [[ ∃x student (x,w) ∧ invited-speaker(y, w)] → pick-up(x,y,w)]]

The facts discussed here are compatible with an indefinite analysis of FCIs. They are

also important because they directly threaten the wide-scope-universal analysis: if FCIs and any

are universals that can scope over a modal like can, as was originally motivated, then why can’t

they raise above the future or other quantifiers in (98) and (100)? The impossibility of this

scoping makes it plausible to argue that in fact FCIs do not scope over can in the original

example (11) after all (see also Dayal 1998:458, footnote 29, for an observation that narrow

scope readings are possible with any).

In the theory of polarity I am arguing for, FCIs are expected not to scope over the

12 Note that indefinite NPs like some generally do not admit bound interpretations, e.g. they are unacceptable

33

nonveridical element that licenses them. It is thus correctly predicted that wide scope readings

will not be possible if they force FCIs to cross the future and therefore end up outside the scope

of their licenser.

3.3 Existential interpretations and quantificational variability of FCIs

As an additional piece of evidence against the universal analysis of FCIs, I review here a number

of constructions where FCIs are interpreted existentially. I illustrate by presenting mostly Greek

examples, but recall again that Spanish and Catalan FCIs behave identically. This

quantificational variability of FCIs, also observed with Q-adverbs, can be explained only if we

assume that FCIs are indefinites.

3.3.1 Imperatives, permissions, and modals:

Permissive imperatives are typical cases where FCIs are interpreted existentially:

(102) a Dhialekse opjodhipote forema.

‘Pick any dress.’

b # Dhialekse kathe forema.

‘Pick every dress.’

c Dhialekse kapjo forema; dhen exi simasia pjo.

‘Pick some dress; it doesn’t matter which one.’

d ! ∃w,x [dress (x,w) pick (you, x,w)]

(103) Spanish (Quer 1999: (3))

Pon cualquer excusa.

put.imper.2sg FC excuse

Give any excuse. (= Give some excuse).

Sentence (102a) is clearly not synonymous to (102b). (102b) is compatible with a situation

where you are asked to pick every dress in a given context, but (102a) is not compatible with

such a situation. Instead, (102a) appears to be synonymous to (102c), inviting us to pick one

dress in a given context-- although it certainly allows us to consider all choices. The same

observation has been made in the literature about any (Horn 1999, 2000); FCIs conform to this

pattern generally, as seen in the Spanish example. The imperatives here are permissive, and

hence can be analyzed as involving an existential modal operator (see discussion in section 4.3);

the examples would then have the reading in (102d). The fact that, unlike with the case of future

just discussed, the FCI sentence here is interpreted existentially confirms the idea that the

with the generic operator: *Some cat hunts mice is not possible with a generic interpretation.

34

quantificational force of the FCI is dependent on the force of the operator that binds it.13

Likewise, FCIs obtain existential readings when construed with permissive and

epistemic modals with the meaning of ‘may’:

(104) Boris na dhialeksis opjodhipote forema.

You may pick any dress.

# You may pick every dress.

(105) (Afisame to fos anameno, ke) Bori na bike mesa opjosdhipote.

(We left the light on, and) Anybody may have come in.

# Everybody may have come in.

Here, again, the FCI-sentences contrast with universals and are interpreted existentially. This

should be attributed to the fact that we have a permission modal, on a par with the permission

imperative. In (105), we have an epistemic possibility modal, may, hence the resulting sentence

again has existential force.

Deontic necessity modals also allow for FCIs; but because the modal operator is a

universal quantifier, FCIs will receive universal interpretations:

(106) a Any minors should be accompanied by their parents.

b ∀w, x [minor (x,w) → ∃y [parent-of (y, x,w) accompany (y,x,w)]]

Likewise, epistemic necessity also induces universal readings, as we see in the following

example from Dayal (1998:457):

(107) a Any pilot must be out flying planes today.

b ∀w, x [[pilot (x,w) plane (y, w)] → fly (x,y,w)]

Commands or instructions also allow for FCIs though, admittedly, the relevant examples are

harder to construct. Here is one example:

13 Universal readings are also allowed in imperatives:(i) Take any textbook in semantics; you will find a mention of Montague in it.

Sentences like the above do not involve permissive imperatives, but can be argued to reflect an underlyingconditional structure, where the imperative clause provides the antecedent. The universal reading is then theresult of binding the FC-variable by the universal quantifier corresponding to the conditional operator. Hence theexistence of universal readings in imperatives of this kind further confirms the observation that thequantificational force of the FCI and any is parasitic on that of the binding quantifier.

35

(108) Context: I am playing a game with a child. I instruct her how to win the game:

Exo kripsi 10 avga se diafora meri. Ja na kerdhisis prepi na vris ena opjodhipote avgo--

dhen exi simasia pjo-- ke na to valis sto kalathi.

‘?I have hidden 10 eggs in various places. Here is how you win: you must find any egg--

it doesn’t really matter which one-- and put it in the basket.’

Opjosdhipote and any are fine in the above situation. (Note that opjodhipote avgo is modified

by the indefinite article). Crucially, what we did was relax the identity requirement and make the

arbitrariness characteristic of free choice a part of the deontic statement itself. Once this is done,

FCIs become licit. A similar example following the same strategy is given below (thanks to

Gregory Ward for judgments and suggestions on constructing such examples):

(109) The hotel manager to a candidate cleaning lady who has just asked him which room to

clean in order to get the job:

‘Dhen exi simasia, to mono pu thelo na dho ine an kseris na katharizis. Pijene tora, ke

katharise opjodhipote dhomatio!’

‘?It doesn’t really matter, all I want to see is whether you know how to clean. Go now

and clean any room (= some room, it doesn’t matter which one)!’

Again, arbitrariness is part of the command and FCIs/any are fine; moreover, they are

interpreted existentially.14 The examples presented here, then, contradict Dayal (1998) who

argues that any is incompatible with commands. In the approach I take here, deontic modals and

commands should in principle allow for FCIs since they are nonveridical, and the fact that we

can find good examples like the ones above is expected.

The fact that it is difficult to find such examples has to do with the additional pragmatic

conditions associated with commands, conditions which may not be entirely compatible with the

semantic content of free choice. For instance, commands seem to depend on precision, hence

quantity and identity may be important, but in FC statements these are precisely the ingredients

that are not important. Likewise, in a potential deontic statement like *You must see any doctor,

it matters whether you must see one doctor or every one (of the contextually relevant ones), and

indeed, it also matters what kind of doctor you see-- if you have a heart problem, for instance, a

14 Additonally, examples can be produced with oposhipote ‘anyhow’:

(i) Prepi na ton dhi oposhipote enas jatros.Literally: A doctor must see him anyhow.(Meaning: One way or another, a doctor must see him).

This use of oposhipote actually seems to necessitate the presence of a deontic modal; a possibility modal is not

36

gynaecologist wouldn’t be an appropriate choice. For a speaker following the Gricean

principles of co-operative conversation, FCIs do not seem to provide the most appropriate

means for occasions where identity may be important. In the case of commands and deontic

modals, then, the occasional unacceptability of FCIs and any presents a genuine case of

infelicity, since the appropriate context can render these felicitous.

3.3.2 Conditionals

In conditionals both universal and existential readings arise and can be clearly distinguished.

Horn 1999, 2000 provides extensive discussion which I will not repeat here. I merely illustrate

that Horn's facts hold for Greek FCIs as well:

(110) a An kimithis me opjondipote, tha se skotoso.

If you sleep with anybody, I’ll kill you.

b An exi lisi opjodhipote provlima, tote tha exi lisi ki afto.

If he has solved any problem, then he has probably also solved this one.

See also Quer 1998, 1999 for the relevant data in in Spanish and Catalan.

3.3.3 Relative clauses

FC-readings of any in relative clause construals (‘subtrigging’ in LeGrand 1975) are claimed to

always be universal. For subtrigging, Greek employs a free relative (FR), as opjosdhipote-items

contain a wh-component:

(111) I’ ll talk to any student I want to {= to whichever student I want to}.

Tha miliso me opjondipote fititi thelo ego!

Essentially, the Greek FC construction corresponds to the wh-ever paradigm in English,

discussed in this context in Horn 1999, 2000. Horn (1999: 26-31) shows that existential

readings are possible with wh-ever. I reconstruct his examples here for Greek, where it is also

obvious that the FCI cannot be interpreted as a universal quantifier:

(112) a Tha pandrefto opjondhipote thelo ego!

‘I’ll marry whoever I want to!

b # I’ll marry every person I want to.

c I”ll marry the person I want to.

so good: #bori na ton dhi oposhipote enas jatros. ‘#A doctor may see him anyhow.’

37

The fact that wh-ever FRs allow for existential readings is also observed in Jacobson 1995. We

revisit subtrigging in section 5.

3.3. 4 Quantificational variability with Q-adverbs

FCIs exemplify quantificational variability characteristic of indefinites when construed with Q-

adverbs (see Quer 1998 and 1999: (41)-(43) for illustrating this point in Catalan and Spanish).

The Q-adverb may be of varying Q-force and the FCI is interpreted accordingly, as the

following examples demonstrate. Q-adverbs are taken to quantify over situations in tripartite

structures of the form below, following Krifka et al. 1995:

(113) a Sinithos dhiavaze i Ariadne opjodhipote vivlio me megali prosoxi.

usually read.3sg the Ariadne FC book with great attention

‘Ariadne usually read any book very carefully.’

b USUALLYs,x [book (x,s) ∧ read (Ariadne, x,s); read-very-carefully

(Ariadne,x,s)]

(114) a Spania dhiavaze i Ariadne opjodhipote vivlio me megali prosoxi.

rarely read.3sg the Ariadne FC book with great attention

‘Rarely did Ariadne read any book very carefully.’

b RARELYs,x [book (x,s) ∧ read (Ariadne, x,s); read-very-carefully

(Ariadne,x,s)]

(115) a I Ariadne dhiavaze panda opjodhipote vivlio me megali prosoxi.

the Ariadne read.3sg always FC book with great attention

‘Ariadne always read any book very carefully.

b ∀s,x [book (x,s) ∧ read (Ariadne, x,s); reads-very-carefully

(Ariadne,x,s)]

In (113) opjodhipote vivlio is bound by usually and is interpreted as most books, in (114) it is

bound by rarely and is interpreted as few books, and (115) expresses a generalization about all

the books Ariadne read.

Note that generic FCIs in subject position can also be construed with Q-adverbs, just

like regular indefinites:

(116) Opjodhipote pagoni ine sinithos megaloprepes; sigekrimena otan aploni tin oura tu.

Any peacock is usually magnificent; specifically, when it speads its tail.

38

(117) Ena pagoni ine sinithos megaloprepes; sigekrimena otan aploni tin oura tu.

A peacock is usually magnificent; specifically, when it speads its tail.

Dayal (1998: 438) claims that any, unlike regular existentials cannot be construed with Q-

adverbs in generic contexts, but the examples above show that this is not true of FCIs, or of any,

for that matter. (The bad examples in Dayal do not contain restrictions for the Q-adverbs, and

this apparently plays a role.).To be fair, the interpretation of FCIs and any in generic contexts

of this type differs from that of regular indefinites in one particular respect: an example like

(116) seems to convey a generalization about all peacocks, whereas the sentence with a regular

indefinite seems to be about most peacocks. This difference is another manifestation of the

plurality associated with FCIs, and we will see in 4.4. that it follows from the fact that FCIs, but

not regular indefinites, force us to exhaust all possible values in epistemic alternatives.

The facts presented in this section are not compatible with the idea that FCIs or any are

universal quantifiers. Instead, they follow naturally if we take it that FCIs are indefinites. This

conclusion carries over to any.

4 The semantics of free choice

We have reached the conclusion that FCIs are indefinites; the question now is: how do FC

indefinites differ from “regular” indefinites?

4.1 The inherent plurality of FCIs: intensionality and variation

Consider the ungrammatical occurrence of a FCI in an affirmative episodic sentence and

contrast it with a grammatical sentence containing a singular indefinite. I assume here following

Heim (1982) that in this case, the variable contributed by the indefinite undergoes text level

existential closure, and is thus bound by a default existential quantifier:

(118) * Opjosdhipote fititis efije.

FC student left.3sg

‘*Any student left.’

(119) Enas fititis efije.

a student left

‘A student left.’

(120) a [[ a student ]] = student(x)

b ∃x [ student(x) ∧ left(x)]

39

The truth of (119) is evaluated in the standard way, i.e. it is true iff there is at least one individual

d ∈ D such that d is the value of x under g, and d is a student and d left. The ungrammaticality

of opjosdhipote suggests that opjosdhipote cannot be interpreted according to this standard

procedure. It is in this sense that a FC indefinite is ‘sensitive’.

A FCI is inherently intensional; its intensionality is marked by the presence of FC

morphology: -dhipote. We can capture this inherent intensionality by postulating that the FC

determiner must combine with an intensional noun phrase (NP) and not an extensional one.

This means that in the standard case the FC determiner must be treated as a type-shifter of type

<<e,t>, <s, <e,t>>>, which, when applied to a property such as the NP denotation, returns an

intensionalized property as its output.

(121) [[ DETFC ]] = λP<e,t>.λw.λx[P(x)(w)]

This way the NP argument is intensionalized, and the FC-phrase, after λ-conversion, ends up

denoting an intensional indefinite which contributes a predicate with two variables instead of

one: one is the regular individual variable that indefinites come with, and the other is a world

variable.

(122) [[ opjosdhipote fititis ]] = student(x)(w)

I use a world variable here, ignoring the differences between worlds and situations (though there

may be reasons that make it preferable to appeal to situations, see footnote 17). I will continue

talking about worlds as a cover term here with this proviso.

In the event that the NP which the FC determiner combines with is already

intensionalized no type-shifting is necessary. This is the case, for instance, with an NP

containing an intensional adjective meaning ‘alleged’: opjosdhipote ipotithemenos sinergatis

ton Germanon ‘any alleged collaborator with the Germans’. In this case the FC determiner

does not alter the type of the NP, since the NP is already intensional. Rather, the FC determiner

functions as a filter passing on the intensional reading to the whole FC-phrase. The type-

shifting possibility we are talking about here, and its optional application, should be understood

as part of the family of type-shifting mechanisms explored in Partee 1987 and related works for

the more familiar class of singular indefinites, bare singulars, and bare plurals.

The difference between “regular” existential indefinites and FC indefinites is thus

reduced to a type difference. Crucially, the world and individual variables of an intensional

indefinite must be bound by an operator that has the ability to bind such variables (a Q-operator,

i.e. a generic, habitual, modal, intensional operator). It is then expected that the text-level

40

existential quantifier cannot bind the individual variable of such an indefinite since it is not a Q-

operator. For the same reason, the world variable of a FCI cannot be bound by a default

existential quantifier over worlds (which would have to assign to the FCI the value of w0, the

actual world). FCIs can be interpreted only if there is some Q-operator in the sentence that can

bind their world variable. In affirmative episodic sentence a FCI is ungrammatical because no

such operator is present, so the world variable of the FCI remains unbound rendering the FCI

uninterpretable and thus ungrammatical. 15

Apart from intensionality, another important lexical semantic feature of FCIs is variation.

The FCI variable must be assigned distinct values in each world or situation we consider. We

can formulate this using the notion of an i(dentity)-alternative, following Dayal 1997 (see also

Giannakidou 1997b):

(123) i-alternatives

A world w1 is an i-alternative wrt iff there exists some w2 such that [[ α]] w1 ≠[[ α ]] w2

Two i-alternatives are worlds w1 and w2 agreeing on everything but the value assigned

to the FCI α. I-alternatives are epistemic alternatives in an obvious way, and naturally, the

worlds that the nonveridical Q operator quantifies over will serve as i-alternatives. In some cases,

i-alternatives may include the totality of the worlds in some individual's epistemic model ME(x)

(w)-- recall that this model corresponds to the doxastic function f (x) (w) = {w': ∀p [s believes

p(w) → p (w')]}, which contains the set of worlds compatible with what x believes in w. This

option explains the readings we get with generic FCIs that we observed in 3.3.4.

(123) gives the criterion for what counts as a valid alternative for FCIs: only worlds

with differing values for the FCI count as i-alternatives. This is an important difference between

FCI variables and regular variables: though the values assigned to the latter may vary from

world to world (but there may be worlds where the value is the same), the values assigned to the

former must vary in each world. To illustrate:

(124) a {A/Some} student may be in danger.

b Opjosdhipote fititis bori na vriskete se kindhino.

Any student may be in danger.

(125) [[ POSSIBLY ∃ x φ ]]w, g = 1 iff there is at least one world w’accessible from w

(wRw’) , such that there is an individual d ∈ D in w’ such that [[ φ ]]w’,g[d/x] = 1.

15 Regarding this idea of a world variable that must be bound, it may be instructive to observe the parallel withbound variables in the pronominal system, i.e. reflexives, and traces under some conceptions.

41

(I return to on the contribution of the modals in section 4.2; for now I just want to consider the

difference between regular indefinites and FCIs.) In interpreting an existential modal statement

like (124a), we consider a set of worlds accessible from w and check whether the existential

statement holds in at least one of these worlds under some value d assigned to x by our

assignment function g. In a model containing three worlds W={ w1, w2, w3} and three

individuals D= {Roxanne, Ariadne, Frank} the statement with a student can be true in the

following state of affairs:

(126) a. w1: g(x)= Roxanne

[[ student(x) ∧ solve (x, this problem) ]] w1,g = 1

b. w2: g(x)= Roxanne

[[ student(x) ∧ solve (x, this problem) ]] w2,g = 0

c. w3: g (x)= Ariadne

[[ student(x) ∧ solve (x, this problem) ]] w3,g = 0

The assignment function may pick the same individual in more than one world, as in the case of

w1 and w2. And there is no requirement that the available values be exhausted: we see that

although D contains a third individual, Frank, this individual is not assigned to x in any world.

A statement with a FCI such as (124b) cannot be evaluated in this situation-- with FCIs, g must

assign a different individual to x in each world, and the available values must be exhausted.

The requirement on exhaustive variation can be treated as a presupposition, and it

explains the quasi-universal readings: as we move from one i-alternative to the other, and as we

consider all alternatives, we exhaust the possible values for the FCI.16 This is encoded in the

definition below:

(127) Free choice item

Let Wi be a non-empty set of possible worlds. A sentence with a free choice item [[ OP

DETFC (P, Q) ]] is true in w0 with respect to Wi iff:

(where OP is a nonveridical operator; P is the descriptive content of the FC-phrase; Q is

the nucleus of the tripartite structure; w0 is the actual world):

a. Presupposition: ∀w1, w2 ∈ W i: [[ α]] w1 ≠[[ α ]] w2

, where α is the free choice phrase.

16 Note that this differs from von Fintel 2000's analysis of ignorance whatever, which posits the existence of i-alternatives as a presupposition. In my definition of FCIs, the existence of possible worlds for the evaluation ofthe FCI is a semantic requirement that FCIs be intepreted on an intensional type. Exhaustive variation furtherrestricts the kinds of worlds that count as appropriate for the evaluation.

42

b.Assertion: [[ OPw,x [P (x,w); Q (x,w)] ]] = 1 where x,w are the variables contributed

by α.

We may additionally incorporate a scalarity requirement: that we consider all relevant i-

alternatives, no matter how remote they may be from what counts as “normal” (cf. Kadmon

and Landman’s widening). Strictly speaking, the operator binds either a situation variable s (e.g.

the habitual operator, Q-adverb, etc), or a world variable w (e.g. the conditional operator, modal

and intensional operators), but we ignore the difference here.

It should be clear in this analysis what the difference is between universal quantifiers

and FCIs. A universal quantifier like every exhausts the values that can be assigned to its

variable in one world. With a FC quantifier, however, the quasi-universal effect is delivered

because we are forced to consider a value in each i-alternatives, and not because we consider all

values in a single alternative. In the assertive component, the FC is bound by the relevant

operator and obtains the quantificational force contributed by that operator.

Let us see now how the proposal works in the core cases we have identified.

4.2 The ambiguity of CAN and the interpretation of FCIs

Recall our original example where any and FCIs seem to be interpreted as universals. Next to

this, consider the other example with can discussed in section 1, where the quasi-universal

reading was not available:

(128) Opjosdhipote fititis bori na lisi afto to provlima. (ability can)

FC-person student can.3sg subj solve.3sg this the problem

‘Any student can solve this problem.’

(129) I epitropi bori na dosi ti thesi se opjondhipote ipopsifio. (possibility can)

The committee can offer this job to any candidate. ≠The committee can offer this job to every candidate.

Modal verbs are quantifiers over possible worlds, hence in modal statements alternatives can be

invoked, and the FCI can be interpreted. The modal alternatives include the modal base which is a

function K specifying the worlds the modal quantifies over (Kratzer 1981, Krifka et al. 1995), and

it is these worlds that serve as i-alternatives. But how do the two different interpretations of the FCI

come about?

The quantificational variability of the FCI reflects the quantificational ambiguity of the

modal can. Ability can, I will argue, is a universal quantifier, but epistemic and permissive can is an

existential quantifier, just like may. I understand that this may seem an unusual move at first glance,

43

but I believe it is justified (see also Chierchia and McConnell-Ginet 1992: 238, for an explicit claim

that the analysis of can as a possibility modal is “certainly not right”; cf. Geurts 1999 for a

different view). I give below a first attempt to define the ability can, following the definition of

necessity modals of Krifka et al. (1995:51); note that at this first approximation the modal base K

includes all possible worlds accessible from w.

(130) Ability CAN (first approximation)

can p is true in a world w with respect to a modal base K and an ordering source <w ("be at

least as normal as") iff:

For all worlds w' in K, there is a world w'' in K such that w''<ww', and for every other

world w'''<ww'' in K, p is true in w'''.

Under this definition, we expect that in every possible world we consider, p will be true in this

possible world. Hence, for an example like John can swim, for each world w we consider, John will

have the ability to swim in that world. Though this definition seems to work fine in worlds where

John knows indeed how to swim (and he knows that he knows that), due to training or natural

talent, in worlds where John didn’t learn how to swim, or hasn’t discovered his natural talent yet, p

is clearly not true. Note that the sentence (128) mirrors this objection: students that didn’t have the

proper math training or haven’t discovered their math talent cannot actually solve this (math)

problem, even if the problem is a very easy one.

The problem here is that our modal base is unconstrained: it includes all possible worlds.

What we need to do is restrict it so that it includes only those worlds in which people have abilities

to do things (because of proper training, natural talents, or whatever other reason), and where people

are aware of these abilities. Let us call these worlds the ability-modal base, and let’s think of it as a

function from w to worlds w’, at least as normal as w, compatible with what an agent x is capable

of doing at w (cf. the parallel to the epistemic models):

(131) Kability (x) (w) = {w’: ∀p [x is capable of p(w) → p (w’)]}

Restricting modal bases this way is not unheard of: Kratzer 1981 postulates epistemic and

deontic modal bases in order to specify the worlds epistemic and deontic modals quantify over, and

to derive the lexical differences between the two. It may be possible that we do not need a distinct

modal base for abilities, and that the ability modal base I posit here can be reduced to an epistemic

modal base, but deciding this is not crucial for the purposes of the present discussion; what counts

is that the ability modal can be analyzed as a universal quantifier. I reformulate (130) in terms of the

restricted ability modal base:

44

(132) Ability CAN

can p is true in a world w with respect to an ability modal base Kability (x) (w) and an

ordering source <w ("be at least as normal as") iff:

For all worlds w' in Kability, there is a world w'' in Kability such that w''<ww', and for every

other world w'''<ww'' in Kability, p is true in w'''.

Once we restrict the worlds ability can quantifies over to worlds where agents have abilities, for the

interpretation of FCIs, we must additionally allow extension of this base to include alternatives

corresponding to less "normal" cases, i.e., even less usual talents and abilities. We need this

extension to get the widening scalarity effect (see also Quer 1998).

Going back to our examples, the assertion of (128) is given below, with the universal

quantifier corresponding to can binding the FCI. The sentence is being interpreted according to the

steps we see in (134):

(133) ∀w, x [[w ∈ Kability ∧ student (x, w)]→ solve (x, this problem, w)]

(134) i. [[ Any student can solve this problem ]] w0,g,Kability = 1 iff ∀w'∈Kability (x) (w),

[[ A student solve this problem ]] w',g =1.

ii. [[ A student solves this problem ]] w',g =1 iff there is at least one individual d∈D such that

[[ student(x) ∧ solve (x, this problem) ]] w',g[d/x] = 1.

iii. Values in i-alternatives

a. i-alt1: g (x)= Roxanne

[[ student(x) ∧ solve (x, this problem) ]] w1,g = 1

b. i-alt2: g(x)= Frank

[[ student(x) ∧ solve (x, this problem) ]] w2,g = 1

c. i-alt3:g (x)= Ariadne

[[ student(x) ∧ solve (x, this problem) ]]w3,g = 1

In the calculation of i-alternatives, the value of the FCI varies from world to world. As we move

from one world to the other, the requirement is that the value assigned to opjosdhipote fititis

‘any student’ will verify student(x) ∧ solve (x, this problem) in each world. Since we pick up

distinct students in each world, and since we consider all worlds included in the extended modal

base, the sentence gives the exhaustive interpretation. It should be clear that this interpretation is

45

the result of exhausting all i-alternatives and of assigning distinct values to the FCIs in each

alternative. Additionally, because we treated the modal can as a universal quantifier over worlds,

the FCI in (128) is interpreted universally too.17

Epistemic and permissive can, on the other hand, are interpreted as existential possibility

modals, and license an existential reading for the FCI. I give here the definition of can as

synonymous to may based on Krifka et al. 1995:51; I abstract away from the differences

between may and might (but we may conjecture that, with might, the worlds included in the

epistemic modal base must be restricted to those worlds which are further away from the ones

that count as “normal”):

(135) Possibility CAN (= may)

can p is true in a world w with respect to an {epistemic/permissive} modal base K and an

ordering source <w ("be at least as normal as") iff it is not the case that must not-p is true

in w with respect to K w <w.

The relevant modal base is referred to as {epistemic/permissive} here, and for the analysis of

permission see Kamp (1978). For our purposes, just the fact that we are dealing with an

existential quantifier in these cases suffices. The interpretation of (129), proceeds as follows:

(136) ∃w,x [[w ∈Kability ∧ candidate (x,w)] ∧ offer-the-job (the committee, x, w)]

(137) i. [[ The committee can offer the job to any candidate]] w0, g,K = 1 iff∃ w'∈K, where K is

the extended epistemic or permissive modal base,

[[ The committee offers the job to a candidate]] w',g =1.

ii.[ The committee offers the job to a candidate]] w',g =1 iff there is at least one individual

d∈D such that [[ candidate (x) ∧ offer-the-job (the committee, x) ]] w',g[d/x] = 1.

(138) Values in i-alternatives

a. i-alt1: g (x)= Ariadne

[[ candidate (x) ∧ offer-the-job (the committee x) ]] w1,g = 0

b. i-alt2: g(x)= Roxanne

[[ candidate (x) ∧ offer-the-job (the committee, x) ]] w2,g = 0

17 Sandro Zucchi indicates a complication here, which points to the direction that we may want to reformulate i-alternatives as situations rather than worlds. The problem is the following. Imagine a world where there are, say,30 students, and one of them solves this problem, and the other 29 do not. The FCI statement should be truedespite the fact that 29 students do not solve the problem. But this is clearly a result that we want to exclude. Ifi-alternatives are chunks of worlds, i.e. situations, rather than worlds, then the problem disappears as we will nothave to consider what happens in the rest of the world containing the 29 students who do not solve the problem.Alternatively, the problem is rectified by quantifying over pairs of worlds and individuals, rather than just

46

c. i-alt3:g(x)= Frank

[[ candidate (x) ∧ offer-the-job (the committee, x) ]] w3,g = 1

Again we see in the calculation of i-alternatives that the value of the FCI varies from world to

world, and again the condition is to consider all alternatives. If the relevant alternatives are the

ones we see in c, then the job will go to either Ariadne, Roxanne, or Frank. This is consistent

with the idea (Kamp 1978) that permissions involve underlying disjunctions of propositions

containing individual terms (or existential quantifiers).

The FCI with the permissive can below is interpreted in a parallel manner (note that

Greek doesn’t have a lexical difference between may and can, and the same verb boro is used in

all cases, as well as for ability). Sentences like the one below say something like: "consider the

books that any book can be assigned as its value in each relevant i-alternative; you are free to

borrow one of those books".

(139) Boris na danistis opjodhipote vivlio.

may.2sg subj borrow.2sg FC book

'You may borrow any book.'

(140) ! ∃w,x [[ w ∈ K ∧ book (x, w)] ∧ borrow (you, x, w)]

(141) i. [[ You may borrow any book ]] w0 , g,K = 1 iff ∃ w'∈K, where K is the extended

permissive modal base, [[ You borrow a book ]] w',g =1.

ii. [[ You borrow a book ]]w' =1 iff there is at least one individual d ∈ D such that

[[ book(x) ∧ borrow (you, x) ]] w',g[d/x]= 1.

iii. Values in i-alternatives

a. i-alt1: g (x)= War and Peace

[[ book(x) ∧ borrow (you, x) ]] w1,g = 0

b. i-alt2: g(x)= the Iliad

[[ book(x) ∧ borrow (you, x) ]] w2,g = 0

c. i-alt3:g(x)= Oedipus Rex

[[ book(x) ∧ borrow (you, x) ]] w3,g = 1

Again we see in the calculation of i-alternatives that the value of the FCI varies from world to

world, and again the condition is to consider all alternatives. Yet the sentence is a permission to

borrow one book and not all, and a situation like the above, where only Oedipus Rex is

borrowed in i-alt3 describes correctly the conditions under which a permissive with a FCI can be

worlds, which is essentially what we do.

47

used appropriately.

A similar analysis can be invoked for the interpretation of FCIs in imperatives:

(142) Dialekse opjodhipote filo; opjo thelis.

pick.imp.2sg FC card; whichever want.2sg

‘Pick any card; whichever you want.’

(143) IMP ∃w, x [[w ∈ Kpermissive ∧ card (x, w)] ∧ pick (you, x, w)]

(144) a. i-alt1: g (x) = ace of spades

IMP [pick (you, ace of spades )]

b. i-alt2: g(x) = queen of hearts

IMP [pick(you, queen of hearts)]

c. i-alt3: g(x) = king of diamonds

IMP [pick (you, king of diamonds)]

Dialekse opjodhipote filo “Pick any card” is an invitation to take some card, be it the ace of

spades, the queen of hearts, or the king of diamonds, etc. but not all of them. The

quantificational force of a permissive imperative can thus be understood as equivalent to that of

permissive modals.

4.3 Episodicity

How does intensionality and exhaustive variation rule out FCIs from affirmative episodic,

negative and interrogative sentences? The common feature of these sentences is that they involve

a single event in an extensional context:

(145) *Idha opjondipote ston kipo.

not saw.1sg anybody in-the garden

(146) *Dhen idha opjondipote ston kipo.

not saw.1sg anybody in-the garden

(147) *Idhes opjondhipote ston kipo?

saw.2sg anybody in-the garden?

(148) ∃!e ∃x [person (x) ∧ saw (I, x, e) ∧ in-the-garden (e)][affirmative episodic]

(149) ¬∃!e ∃x [person (x) ∧ saw (I, x, e) ∧ in-the-garden (e)] [negation]

(150) ? ∃!e ∃x [person (x) ∧ saw (you, x, e) ∧ in-the-garden (e)] [interrogative]

This means that the second clause of our licensing condition (60) is not obeyed. Recall:

48

(151) Licensing condition on FCIs

A FCI α is grammatical in a sentence S iff:

(i) α is in the scope of a nonveridical operator β; and

(ii) S is not episodic.

But now that we have established intensionality and variation as the lexical semantic ingredients

of FC, we can predict directly that episodic past will always block the possibility of invoking i-

alternatives: since we are dealing with one event, the identity of the participants is fixed and

cannot vary. And, of course, episodic sentences of this type are extensional and thus provide no

worlds that can serve as i-alternatives-- unlike intensional episodic sentences which do so. So

we can actually dispense with both conditions above because we made them derivable from the

intensional and variational semantics of FCIs.

For the sake of completeness, let us consider two more cases. As expected under the

present analysis, FCIs are fine with individual-level predicates (where e is assumed to be bound

inherently by GEN as in Chierchia 1995, or absent altogether as in Kratzer 1995):

(152) I Ariadne gnorizi opjondipote sto tmima.

the Ariadne knows FC-person in-the department

Ariadne knows anybody in the department.

(153) [[ know]] =λx1 λx2 GENs (in (x1,x2,s)) [know(x1,x2,s)] (Chierchia 1995)

If i-level predication allows us to invoke alternatives, like the ones below, then sentence (152)

tells us that Ariadne stands in the inherently pluralized generic relation know to individuals that

can be assigned as values to the FCI in each of the relevant situations we consider.

(154) a. i-alt1: g(x)= Roxanne

[[ person (x) ∧ know (Ariadne, x) ]] w1,g = 1

b. i-alt2: g(x)= Ariadne

[[ person (x) ∧ know (Ariadne, x) ]] w2,g = 1

c. i-alt3:g(x)= Frank

[[ person (x) ∧ know (Ariadne, x) ]] w3,g = 1

Finally, let us consider the ungrammaticality of FCIs in the scope of veridical attitudes,

i.e. epistemic, dream, and factive attitudes. Since propositional attitudes do provide worlds that

could serve as i-alternatives, how come epistemic attitudes are incompatible with FCIs? The

49

answer to this question lies in the fact that in these cases, although we do have intensionality, the

variation requirement is not satisfied. Consider the case of a factive first:

(155) * I Ariadne metaniose pu idhe opjondipote filo tis.

‘Ariadne regrets that she saw any friend of hers.’

The factive complement is veridical episodic. This means that the FCI will be forced to receive

the same value in all relevant i-alternatives, and as a result, it will be ruled out. (There are two

possible sets of alternatives here, the speaker’s or Ariadne’s epistemic model, but I consider

here only the former):

(156) [[ I Ariadne metaniose pu idhe opjondipote filo tis]] ME(s) = 1 →∃!e ∃x [friend (x) ∧ saw (Ariadne, x, e)], thus

∀w, w'∈ ME (s), [[ opjondipote filo tis]] w = [[ opjondipote filo tis]]

w’

Exactly the same thing can be said for sentences under epistemic, dream/fiction verbs,

and the rest of the directive intensional class (with adjustments regarding the model of

evaluation, see Giannakidou 1999). I illustrate here the epistemic and dream/fiction case; the

relevant alternatives are the worlds corresponding to Ariadne’s epistemic model:

(157) a * I Ariadne pistepse oti idhe opjondipote filo tis.

‘*Ariadne believed that she saw any friend of hers.’

b [[ I Ariadne pistepse oti idhe opjondipote filo tis]] ME(Ariadne) = 1 →∃!e ∃x [friend (x) ∧ saw (Ariadne, x, e)], thus

∀w, w'∈ ME (Ariadne), [[ opjondipote filo tis ]] w = [[ opjondipote filo tis]]

w’

(158) a * I Ariadne onireftike oti idhe opjondipote filo tis.

‘* Ariadne dreamt that she saw any friend of hers.’

b [[ I Ariadne onireftike oti idhe opjondipote filo tis ]] ME(Ariadne) = 1 →∃!e ∃x [friend (x) ∧ saw (Ariadne, x, e)], thus

∀w, w'∈ ME (Ariadne), [[ opjondipote filo tis]] w = [[ opjondipote filo tis]]

w’

Recall that Greek FCIs are not unique in ruling out FCIs in the scope of veridical attitudes;

Spanish and Catalan FCIs behave on a par, as we saw in 2.2, and likewise any.

4.4 Generic FCIs and Q-adverbs

Recall now the examples in 3.3.4 which illustrated that FCIs are interpreted in a different way

50

from regular indefinites when occuring as generic subjects. I provide here two new examples.

(159) Opjosdhipote fititis tu Yale meleta sinithos Shakespeare to proto etos.

Any Yalie usually studies Shakespeare in the first year.

(160) Enas fititis tu Yale meleta sinithos Shakespeare to proto etos.

A Yalie usually studies Shakespeare in the first year.

We said that the interpretation of FCIs and any in generic contexts of this type differs from that

of regular indefinites in that the FCI statement seems to convey a generalization about all Yalies.

The generic sentence with a regular indefinite does not seem to have this reading.

In our terms, the 'consider all' reading arises because FCIs carry the presupposition that

we consider all alternatives. Singular indefinites simply do not carry this presupposition. I

illustrate here the interpretation of the FCI sentence; the meaning of the sentence with the

regular indefinite is only the component in b, hence a Yalie is interpreted as most Yalies.

(161) [[ (159) ]] is true in wo iff:

a. ∀ w1 , w2

∈ ME (speaker): [[ Yalie (x)]] w1 ≠[[ Yalie (x) ]] w2 (presupposition)

b. USUALLYs,x [Yalie (x, s) ∧ be-first-year (x, s); study (x, Skakespeare, s)]

(assertion)

Crucially, i-alternatives are drawn from the speaker's epistemic model, which includes of course

the worlds the Q-operator ranges over, while the any/FCI variable is bound by the Q-operator.

This allows us to consider all values for any Yalie but actually assert a generalization about

most of Yalies.

Both sentences differ from the one below, with the universal quantifier, which just picks

the extension of Yalies in the actual world:

(162) a Every Yalie usually studies Shakespeare in his first year.

b [[ (162a) ]] is true in wo iff:

∀x [[Yalie (x, wo ) ∧ be-first-year (x)] →USUALLYs [study (x, Shakespeare, s)]

A similar contrast presented in Dayal (1998: (9c,d)) as an argument in favor of a

universal analysis of any, has the same explanation:

(163) a Opjosdhipote filosofos kani lathos kamia fora.

Any philosopher is sometimes wrong.

51

b Kathe filosofos kani lathos kamia fora.

Every philosopher is sometimes wrong.

The universal statement has the logical structure we see in (164) and can be interpreted as true

or false in the actual world w0. But the statement with the FC is evaluated against the

presupposition and the assertion in (165), and in order for it to be true or false, we have to

consider all i-alternatives. The two are clearly not equivalent.

(164) [[ (163b) ]] is true in wo iff:

∀x [philosopher (x, wo) → ∃s [be-wrong (x, s)]

(165) [[ (163a) ]] is true in wo iff:

a. ∀ w1 , w2 ∈ ME (speaker): [[ philosopher (x)]] w1 ≠ [[ philosopher (x) ]] w2

(presupposition)

∃s,x [philosopher (x, s) ∧ be-wrong (x, s)] (assertion)

Hence, just like in the previous example, the quasi-universal effect is a by-product of the fact

that i-alternatives include not only the situations contributed by the existential operator binding

the FCI, but the totality of the epistemic alternatives of the speaker.

To sum up, I proposed in this section a semantics for FCIs which straightforwardly

derives the universal-like readings and the limited distribution of FCIs in nonveridical and

nonepisodic contexts from their lexical semantic properties of intensionality and exhaustive

variation. Most importantly, this semantics makes it unecessary to postulate additional licensing

conditions. The incompatibility with extensional veridical episodic contexts was shown to arise

simply as a clash between the lexical semantics of FCIs and the context of occurrence resulting

in uninterpretability. No recourse to universal quantification was necessary; in fact we saw that

appealing to universality would give us the wrong results in many cases. In the next section, I

consider the issue of subtrigging and show that there is no need to appeal to universality for this

case either. Finally, we consider the consequences of this analysis for any.

5 Revisiting any

We are now in a position to return to one of the questions we set out to answer at the beginning:

if we can show that FCIs are not universal quantifiers but indefinites, what implications does

this have for the analysis of any?

Dayal (1998) presents a recent revival of the two anys view. She assumes that there are

two anys, a FC one and a polarity sensitive (PS) one. The latter is an existential licensed in

52

downward entailing contexts, an assumption known to be problematic (Giannakidou 1997,

1998, 1999, Zwarts 1995; see also discussion in Ladusaw 1996)-- though, since "PS" any will

not be a concern here, I will not review the arguments against it. FC-any is an intensional

universal quantifier whose distribution is additionally regulated by contextual vagueness. (Note

that the adopted opposition "PS" vs. FC is misleading: under the more fine-grained approach to

polarity phenomena I am pursuing, FC is an instance of polarity.).

Since I have been at pains to show that there is no evidence that FCIs are universals, the

view of FC-any as a universal can no longer be maintained, unless there is strong independent

evidence supporting it. The relevant literature, however, presents no such evidence-- we saw that

the most popular arguments for universality, namely almost/absolutely modification and

exceptive phrases, are unreliable and have alternative explanations compatible with the idea that

FCIs are indefinites. Dayal (1998) suggests that subtrigging can indeed be taken as evidence

that any is a universal. I consider this issue here and argue, following Quer 1998, 2000, that

what we see as universality is in fact the result of an underlying conditional structure where the

any variable is bound by the universal quantifier contributed by an implicit conditional operator.

The idea that subtrigged structures involve underlying conditionals actually goes back to

LeGrand's (1975) original analysis of subtrigging.

If subtrigging does not require universality, the split between two anys, one existential

and one universal, seems at least redundant. The only remaining issue, then, is whether we need

additionally contextual vagueness in order to account for the distribution of any. We see that we

do not. The alleged workings of contextual vagueness can all be attributed to nonveridicality and

variation.

5.1 Subtrigging

The term 'subtrigging' is used to refer to cases where any and by extension FCIs appear

grammatically followed by a relative clause, giving rise to universal-like readings. Such

construals appear in contexts where any and FCIs would be grammatical anyway, e.g. under the

generic/habitual operator and i-level predicates with imperfective aspect, or in an otherwise

hostile environment, e.g. episodic perfective past.

(166) FCI-friendly environment

a Ekino to vradi, o Janis miluse me opjandhipote jineka (epidi itan poli monos).

that the night, the John talked.impf.3sg with FC woman

That night John talked to any woman (because he was lonely).

b Opjosdhipote su to ipe afto ine trelos.

FC-person you it said.pf.3sg this is isane

Whoever told you this is insane.

53

(167) FCI-hostile environment

a *John talked to any woman.

b Ekino to vradi, o Janis milise me opja(?dhipote) jineka ton plisiase.

that the night, the John talked.perf.3sg with FC woman him approached.perf.3sg

That night John talked to any woman who came up to him.

c John talked to whoever came up to him.

(168) Opjos(?dhipote) anakalipse miga sti supa tu, dhen efaje.

FC-person discovered.perf.3sg fly in-the soup his, not ate.perf.3sg

Whoever discovered a fly in his soup didn't have dinner.

In FCI-hostile contexts, the relative clause seems to 'save' an otherwise ungrammatical sentence.

Similar constructions are possible with FC free relatives in Spanish and Catalan (see Quer 1998,

1999, 2000 for detailed descriptions). In Greek, as we see, we also have a free relative

construction, comparable to the English whoever. We also see that in FCI-hostile environments,

the FC option is slightly degraded compared to the regular free relative opjos. Degrading may

amount to complete ungrammaticality in certain cases, as we see below:

(169) *Opjosdhipote su to ipe afto pethane.

FC-person you this said.perf.3sg died.perf.3sg

* Whoever told you this died.

(166b) with the stative i-level predicate is insane and (167b), (168) with potentially iterative

predicates, contrast minimally with (169) which contains the once-only predicate died. The

contrast obtains in English as well-- the * indicates the unacceptability of (169) with the quasi-

universal subtrigged reading; Quer 1998, 1999 reports similar contrasts in Catalan and Spanish.

The English translation of (169) may indeed be acceptable as an identity statement, but with an

existential reading: The person who told you this died. This reading, labelled in Tredinnick 1996

the 'don’t know' reading, is clearly not an option with FCIs in Greek, Catalan, and Spanish.

In the examples above, any and FCIs seem to be interpreted universally, which motivated

Dayal's (1998) account of any as a universal quantifier. Yet we saw in section 3.3.3 that it is

possible to have subtrigged FCIs with existential readings. The data are repeated here:

(170) a Tha pandrefto opjondhipote thelo ego!

‘I’ll marry whoever I want!

b # I’ll marry every person I want.

c I”ll marry the person I want.

54

So at least for FCIs and -ever we cannnot maintain that subtrigged construals are always

equivalent to universal readings. But the universal-like readings are indeed a reality in the

majority of cases, and the presence of the relative clause is crucial in licensing them. Let us

consider now Dayal's explanation.

5.1.1 Dayal (1998)

To see how subtrigging is explained, consider first the simple paradigmatic case. Dayal

proposes that any is a universal quantifier "whose domain of quantification is the set of possible

individuals of the relevant kind, rather than a set of particular individuals" (Dayal 1998: 447).18

This is represented by the universal quantifier binding the situation index on the description of

the any-phrase. In a simple case with any construed with the generic operator, we would have

the following structure (Dayal 1998:448):

(171) a Any owl hunts mice.

b ∀s,x [owl (x,s) ∧ C(s)]

GEN s' [s<s' ∧ C'(s')] ∃y (mice (y,s') ∧ hunt (x,y,s')]]

Here both the universal quantifier any and the generic operator contribute tripartite structures

and genericity comes from two sources: (a) from any itself since it introduces a situation

variable, and (b) from the matrix predicate, as usually assumed for generic VPs. The formula

says that all situations s that have an owl in them generally extend to situations in which an owl

hunts mice. C(s) represents a condition that we exclude non-normal cases of owls, e.g. sickly

owls, newborn owls, etc. which would constitute exceptions to the generalization expressed by

the generic sentence.

An immediate problem is why the genericity inherent to any does not suffice to make

any grammatical in the absence of VP genericity, as is the case with episodic past:

18 Dayal (1998: 442-445) also claims that any is necessarily construed with essential properties, but this is notentirely accurate:

(i) Opjoshipote fititis stin taksi tis Marias etixe na psifisi to sindiritiko koma.‘?Any student who was in in Mary’s class happened to vote for the conservative party.’

(ii) I Maria diavase opjodhipote grama etixe na vriskete pano sto grafio tis.‘?Mary read any book which happened to be on her desk.

(iii) a. I {will/promise to} read any book which happens to be on my desk.b I’ll eat any food you {happen to/decide to} cook for me.

The Greek examples here are parallel to the ones given in Dayal. FCIs are fine with etixe na ‘happened to’,

55

(172) a *Yesterday, any owl hunted mice.

b Yesterday, every owl hunted mice.

In order to rule out these cases, Dayal resorts to a pragmatic explanation which I criticize in

detail in 5.3. The idea is that in episodic sentences we have no way to restrict the any-

quantification, because we quantify over all possible situations containing owls. Consider now

the subtrigged example (167b). This sentence has the logical form below (Dayal 1998: 452):

(173) ∀s,x [woman (x,s) ∧ C(s) ∧ ∃s'' [s<s'' ∧ P(s'') ∧ (came-up (x,j,s'')]] ∃s' [s<s' ∧that-night (s') ∧ talk (j,x,s')]

The common noun together with the relative clause restrict quantification to possible woman

situations which are temporally anchored to the interval provided by the relative clause. It is this

property of temporal anchoring that makes any acceptable with a relative clause.

5.1.2 Subtrigging as an underlying conditional

In the indefinite approach that I am pursuing here there is an alternative account that can explain

the subtrigged facts just as well. Quer 1998, 2000 proposes an analysis of subtrigging as an

underlying conditional structure, where the content of the relative clause functions as the

restriction of the implicit conditional operator. In this account, a sentence like (167b) will have

the structure schematized in (174):

(174) ∀w, x [[woman (x, w) ∧ came-up (x,j,w)] → talk-to (j,x,w)]

The only universal quantifier here is provided by the postulated conditional operator; the any

phrase is an indefinite bound by that operator. As a result, subtrigged any phrases are

interpreted universally. In this account, the presence of a relative clause is crucial in that it

provides the restriction of the conditional.

This analysis carries over directly to the FC-cases under consideration. Crucially, the

availability of a conditional structure hinges on whether the sentence containing FCIs will

provide possible worlds or situations that can function as i-alternatives. When the sentence

provides such alternatives, as is the case with generic/habitual and stative sentences with

imperfective aspect, then the conditional structure is licensed. But in the absence of genericity,

the worlds have to come from somewhere else: from the iterativity of the predicates (especially

the matrix), a point also emphasized in Dayal. Iterative predicates allow the creation of the

which forces a non-essential property reading; any is fine too. The examples in (iii) are from Horn (1999:10).

56

conditional structure, as we witness in (167b), (168); once-only predicates like die, on the other

hand, cannot feed conditional structures and FCIs are thus ruled out. I won't go into further

details here, for discussion see Quer 1998, 2000.19 Nonveridicality is satisfied in these cases

because the protasis of conditionals, where any/FCIs occur, is nonveridical (Giannakidou 1997).

5.2 Any and (non)veridicality

Recall the data discussed so far and summarized in Table 1. Just like FCIs and APIs, any is

ruled out from veridical contexts. But there are two differences. First, unlike APIs and FCIs, any

may be licensed also in a negative factive complement:

(175) a Lucy regrets that she talked to anybody.

b * Lucy is glad that she talked to anybody.

Any in this case is licensed by a negative implicature (Linebarger 1980). The Greek PIs, which

are licensed by nonveridicality, cannot be grammatical in a veridical context such as the

complement of a factive verb, as we saw in section 2.

Second, any is ungrammatical in some nonveridical contexts: perhaps clauses,

disjunctions, and partially in the scope of strong intensional predicates, as illustrated below:

(176) a Isos o Pavlos na milise me {opjondhipote/kanenan}.

perhaps the Paul subj talked.3sg with FCI/ API

‘*Perhaps Paul talked to anybody.’

b I bike {opjosdhipote/kanenas} mesa i afisame to fos anameno.

or entered.3sg FCI/ API in or left.1pl the light lit

‘*Either anyone came in or we left the light on.’

(177) a ‘*I want you to buy any book.’

b ‘I insist you allow anyone in.’

The contrast we observe here suggests that any and the Greek PIs, though related, they arguably

form distinct paradigms. So what is the correct distributional constraint for any? I will argue,

adopting Giannakidou 1999, that unlike APIs and FCIs, any is anti-licensed by veridicality.

Anti-licensing is the weaker form of semantic dependency corresponding to a negative relation

between PIs and veridicality. It says that in order to be grammatical a PI must not be in the

19 Quer actually tries to reduce all cases of subtrigging, even the ones in episodic contexts, to genericity. Hisgoal is motivated by the observation that subjunctive free relatives, which are the counterpart of subtrigged FCin Catalan and Spanish, require imperfective aspect in all cases. This doesn’t hold for Greek, as we see in theexamples discussed here, so a reduction of subtrigging to genericity would be unnecessary, at least for Greek.

57

scope of a veridical operator (recall the discussion in section 2). The theory of polarity I assume

here allows for this option, and any looks like a pretty good candidate for using it. The negative

dependency of any does not allow the strong prediction that it appear in all nonveridical

environments, although it surely raises the expectation that it will appear in most of these. But it

is quite feasible that there will be some nonveridical contexts in which any will not be licit, as is

the case.

We can then formulate the following condition for any:

(178) Condition regulating the distribution of any

i. Any will not be grammatical in a sentence S if any is interpreted in the scope of a

veridical expression in S.

. ii. In certain cases, clause i can be voided if S gives rise to a negative implicature.

If the veridical expression is embedded under a nonveridical operator, then any is located in the

global scope of the nonveridical operator and will therefore be grammatical, as in John didn't

say that he talked to anybody (recall also fn.5). Intervention effects with quantifiers (for which

Linebarger's (1987) immediate scope constraint is invoked) further pose constraints of a

different kind, but I will not discuss them here for reasons of space.

As with FCIs, ultimately, we would like to derive the constraint in (178) from the lexical

semantics of any. Although space prevents a detailed analysis here, we can address the obvious

question: can we treat the any phrase as an intensional indefinite just like FCIs? If we want to

treat any as a unitary item, which is the desired option since we have no evidence from English

to the contrary, the answer must be negative.

We saw that any is fine in extensional contexts with episodic negation and in questions

simply because these are not veridical contexts. This suggests that intensionality and variation

are not encoded in the semantics of any the way they are in the semantics of FCIs. We can still

argue, however, that any is an indefinite. Additionally, because any is ungrammatical in an

extensional context in the absence of negation, we can assume that, just like FCIs, any cannot

have its variable bound by a default existential quantifier provided by the context. This is any's

lexical 'deficit', so to speak, responsible for restricting its distribution. Any is not alone in having

this deficit; in Giannakidou (1998) I identified other members of this class of dependent

indefinites-- the nonemphatic APIs kanenas, tipota, etc. were shown to be dependent in this way

too. Dependent indefinites cannot assert existence like regular indefinites do: they cannot

undergo existential closure at the top text-level. In a negative episodic sentence, however, or in a

question, existential closure can apply because there is some nonveridical operator above the

existential at the top level, i.e. negation and the question operator, thus suspending the

possibility of asserting existence. Dependent indefinites and any are therefore fine under

58

negation, the question operator and other nonveridical operators because of this reason.

We can summarize the lexical characteristics of any in the following:

(179) Any

a. Any P is an extensional indefinite of the form P(x), where x is an individual variable.

b. The x variable is dependent: it cannot be bound by a default existential, unless there is

another nonveridical operator above the existential. If the nonveridical operator is a Q-

operator, then the Q-operator binds the x variable, as is standardly the case with

indefinites.

c. It is conversationally implicated that there are i-alternatives such that:

∀w1, w2 [[ α]] w1 ≠[[ α]] w2, where α is the any P.

(As with FCIs, we can augment (179) with scalarity: we consider all relevant i-alternatives, no

matter how remote they may be from what counts as “normal”, cf. Kadmon and Landman’s

widening, and Lee and Horn's 1994 analysis of any as an indefinite plus even.). Here we

postulate exhaustive variation as a weaker requirement on any, i.e. as a conversational

implicature instead of presupposition as we did for FCIs, and it is responsible for the FC

readings of any in nonepisodic contexts. In extensional contexts, e.g. with episodic negation and

questions, there are no worlds that can serve as i-alternatives, and the implicature is cancelled.

The implicature of exhaustive variation is not part of the lexical semantics of the API-indefinites

kanenas, tipota, etc. which is why these never give rise to free choice readings.

The discussion here is certainly far from being a detailed analysis of any, but, given the

facts presented in this paper, it seems fair to say that it can serve as a relatively solid basis for

such an analysis. The important finding is that a treatment of any as a dependent indefinite

provides a viable way to explain its distribution without making it necessary to appeal to two

anys.

I consider next the issue of contextual vagueness.

5.3 Against contextual vagueness

Let us consider now Contextual Vagueness (Dayal 1998: (57)):

(180) a Contextual Vagueness: any is only appropriate in contexts where the speaker

cannot identify the individual or individuals who verify p.

b Revised Vagueness Requirement: Any (A) (Op B) is felicitous iff A « B is not

contextually salient in any relevant world; where Op may be possibility, necessity, !, the

reverse or null.

59

Note that the revised vagueness requirement in b appeals to modality, suggesting that contextual

vagueness alone cannot be used as a licensing principle for any. Dayal explicitly rejects the

relevance of notions like nonexistence (a notion which Dayal, erroneously, uses as equivalent to

nonveridicality) in licensing any (Dayal 1998:442).

Hoewever, the contextual vagueness condition cannot afford the empirical coverage of

the (non)veridicality based (178). We can confirm this by examining first cases that (180) is

supposed to rule out: we see that the exclusion of any in these cases can have an alternative

explanation in terms of veridicality. Then we will consider cases where (180) seems to make

wrong predictions. It will be shown that contextual vagueness wrongly predicts that any would

be fine under believe, dream, etc, since (180) is satisfied in these contexts, contrary to fact. It is

also predicted that whenever identity is unknown or irrelevant any will be acceptable, again

contrary to fact. The obvious conclusion will be that (180) cannot be neither a necessary nor a

sufficient condition for the licensing of any.

5.3.1 Affirmative veridical sentences

The first indication towards this conclusion comes, surprisingly, from Dayal herself. Consider

the ungrammaticality of any in veridical affirmative sentences:

(181) *Yesterday, John talked to any woman.

Crucially, Dayal does not use contextual vagueness to rule out any in this core case. This is so

because in affirmative veridical sentences contextual vagueness may indeed be satisfied: a

speaker may utter (181) without knowing the identity of the particular woman or women that

John talked to. In fact, we can modify (181) as (181’), by adding a clause establishing that we

don’t know the identity of the women, but this addition does not affect the grammaticality of

any:

(181’) *Yesterday, John talked to any woman, but I have no idea {who they were/who it was}.

Contextual vagueness (and comparable notions like Tovena and Jayez’s 1997 arbitrariness)

actually makes the prediction that in contexts like the above, where it is made explicit that

identity is unknown, any should be grammatical, but this prediction clearly fails.

Since contextual vagueness won’t help, Dayal appeals to a more radical pragmatic

explanation for the ungrammaticality of (181). It is claimed that “there will be many situations

that will render this statement false, for example all the those women situations that do not

overlap with John's existence", and "there is something infelicitous in making a statement that is

doomed to be false” (Dayal 1998: 453). So a sentence like (181) is ungrammatical due to "a

60

presupposition failure: in using an any phrase the speaker chooses explicitly to talk about all

possible situations but in making an assertion about a bounded time interval, she must focus on

a restricted set of situations" (ibid.).

There are two flaws in the reasoning here. First, how come presupposition failure leads

to ungrammaticality? We know of other cases of presupposition failure which do not cause

ungrammaticality: if I say “bring me the red dress” in the absence of a unique dress, I am just

making an infelicitous utterance, but I am not producing an ungrammatical sentence.

Moreover, one can still utter the sentence below, which indeed invites us to consider all

possible woman situations, but the sentence is by no means ungrammatical; it just leads us to

assume that the speaker is ignorant of some basic physical properties of the universe “doomed

to falsehood”-- to use Dayal's phrasing-- in the (actual) world, where women have lived before

John’s birth:

(182) Yesterday, John talked to every woman you can possibly imagine.

Why is (182) grammatical? The reasoning applied to any should apply to this sentence too, and

we would expect to get exactly the same presupposition clash we presumably get for any. Yet

the sentence is not ungrammatical, and the speakers are able to apply some contextual restriction

and interpret every woman. Why can’t this be done in the case of any?

Second, how come a statement doomed to be false is ungrammatical? One can imagine

lots of situations where one knows that what one says is probably doomed to be false, but this

doesn’t affect the grammaticality of the produced sentence. Lying is a prototypical case where

we say something we actually know is false, but a liar’s sentence is not ungrammatical, it is not

even infelicitous. It is simply false.

Note, finally, that if we tried to fix the context in such a way so as to guarantee that we

consider only women whose existence overlaps with John (thus satisfying the requirement on

temporal anchoring), the crucial sentence would still be ungrammatical:

(183) (John is participating in a contest)

*There were rumors that 10 women would participate in the contest and that they would

stay at the same hotel with John. By 10 pm yesterday, John kissed any of them.

By Dayal's explanation, we would expect an improvement in this sentence, since (a) now we

exclude the woman situations which do not overlap with John's existence and we only consider

the woman situations which overlap with John’s participation in the contest; and (b) contextual

vagueness is obeyed: we have no idea who the women were, or if they indeed arrived at the

hotel. The occurrence of any, however, is still not good.

61

What we are witnessing here is the failure of excluding PIs on purely pragmatic

grounds, and this failure plagues not only Dayal, but any approach that would treat polarity

phenomena as part of the discourse grammar alone. The general problem is that licensing

failures result to ungrammaticality and cannot be fixed by the context, as we would expect if the

source were pragmatic.20 This conclusion, of course, is not meant to imply that pragmatic

conditions will not play a role; in fact, we did allow pragmatic factors to bear weight in our

discussion (recall the case of deontic modals and commands (3.3.1), as well as the

presupposition of exhaustive variation). But leaving room for pragmatic factors in addition to

the semantics ones is very different from reducing polarity to felicity. There is enough evidence

crosslinguistically to ensure that such a reduction does not do justice to the empirical data.

5.3.2 Arguments for contextual vagueness which are arguments for (non)veridicality

We see now that occurrences of any that Dayal rules out by contextual vagueness are also

excluded by veridicality.

(i) Contrast between every and any.

Dayal presents examples showing that every, but not any, will be OK if the set referred to by the

20 A reviewer suggests examples like the following as indications that any may be fixed by the context andappear felicitously even in episodic sentences:

(i) Context: There was a lot of food at our party last night but we have nothing left. People ate a lot. Theyalso took food home.(a) And any left-overs were thrown out.

Note, though, that sentence (ib), unlike (ia) above, is still ungrammatical in this context:

(b) *And any chips were thrown out.

(ib) differs minimally from (ia) in that the any phrase does not contain a deverbal noun: left-overs in (ia) butchips in (ib). The deverbal noun can provide a clausal structure, and the acceptability of (ia) relies crucially onthis, a fact indicating that we are dealing with subtrigging. The availability of clausal structure may sanction anyeven in the absence of a deverbal noun, as in the sentences below, brought to my attention by Larry Horn:

(i) c And any crumbs were vacuumed away.(ii) We straightened out the lawn furniture and cut down the saplings. And any {weeds/*tulips} were mowed

down.

In both cases, the any-phrase seems to be interpreted in a clausal structure of the form 'any remaining crumbs'and 'any remaining weeds'. Why this clausal structure is licensed with crumbs and weeds but not with chips andtulips must probably be attributed to the fact that the existence of crumbs and weeds is predictable in therespective contexs, but that of tulips and chips is not, and this allows for the clausal interpretation of any. Atany rate, the grammatical cases under consideration here can be understood as cases of subtrigging and provide noargument that the context alone can affect the grammaticality of any. The context, or a deverbal noun can enableclausal intepretations for any, but in the absence of clausal structure internal to the any-phrase, any remainsungrammatical.

62

quantifier has been previously introduced in the discourse:

(184) There were twenty students at the lecture and {every/*any} student who was there said

it was inspiring.

In a context like (184), there were indeed students at the lecture, which makes the sentence

veridical and hence rules out any. Similar examples with every are discussed in Giannakidou

1999. Likewise, the following sentence (Dayal’s (64)) is also ruled out by veridicality, since

Susan found the books she was looking for:

(185) * Susan found any book she had been looking for at Borders. And what’s more, they

were on sale! 21

(ii) Existential constructions

As we saw in 2.1, any is ungrammatical in existential constructions:

(186) a There is{every/*any book} by Chomsky in this library.

b There is {everything/*anything} Mary had asked for in this store.

Dayal argues that “in there-insertion contexts the relevant set is made contextually salient”, but

this is too weak. We can utter for instance the sentence below, satisfying vagueness, but the

sentence is still ungrammatical with any:

(187) There was {everything/*anything} Mary had asked for in this store, but I have no idea

what that was.

For veridicality the explanation seems straightforward: existential contexts are veridical

and hence inappropriate environments for PIs.

5.3.3 Propositional attitudes

Dayal’s account predicts that any should be sanctioned with all propositional attitudes, contrary

to fact:

(188) a John would like us to buy any book on this list for his birthday.

b John told us to buy any book on this list for his birthday.

21 Note that *Susan found any book she had been looking for at Borders is also ungrammatical because ofveridicality. Subtrigging cannot work in this case because a predicate like look for is not iterative or inherentlygeneric, hence it cannot create the necessary conditional structure for subtrigging.

63

c * John believes that we bought any book on this list for his birthday.

d * John dreamt that we bought any book on this list for his birthday.

Recall from section 2 that FCIs crosslinguistically exhibit this contrast. In propositional attitude

contexts contextual vagueness is satisfied: there is an intensional operator as required by

Dayal’s (180), and neither the speaker or the hearer have any idea what book(s) will be bought

or have been bought, if any. There is a clear constrast, however, between the sentences in a,b,

with nonveridical attitudes which allow for any, and those in c,d with veridical verbs which rule

out any. The ungrammaticality suggests that, for any, just any kind of modality is not enough.

In a theory based on nonveridicality the contrast is expected: as with other PI-paradigms,

veridical modalities are not compatible with any.

Contextual vagueness, then, does not seem to be a necessary or sufficient condition for

any. What can be misleadingly read off as workings of contextual vagueness are in fact results

of nonveridicality. To be sure, there is a connection between nonveridicality and vagueness, in

the sense that the former guarantees the latter due to suspension of truth and existence. But if

vagueness is satisfied in a veridical context, as in the cases of veridical intensional attitudes, it is

nonveridicality that matters for licensing any and not contextual vagueness.

6 Conclusions

The primary goal of this paper has been to link the constraints on the distribution of free choice

items to their semantics, as part of the more general argument that it is lexical sensitivity that

determines limited distribution in polarity items. The sensitivity of free choice items was treated

as a type difference between “regular” indefinites, which do not exhibit limited distribution, and

free choice indefinites, which are only admitted in nonveridical and nonepisodic contexts.

Unlike regular indefinites, free choice items are intensional: they come with a world variable that

must be bound by some operator that has the ability to bind such a variable; if such an operator

is not present, the world variable remains unbound, rendering the free choice item

uninterpretable and thus ungrammatical. Additionally, the value assigned to a free choice item

must differ in each alternative (world or situation) we consider. These lexical semantic features

of free choice items are responsible for restricting their distribution to nonveridical and

nonepisodic contexts: only in such contexts are both features satisfied.

This analysis has two important consequences. First, it supports the idea that limited

distribution phenomena covered by the term “polarity” involve, one way or another, sensitivity

to veridicality and nonveridicality or subproperties thereof. Second, by deriving the restrictions

on their distribution from the lexical semantics of free choice items, we have established that at

64

least some of what are traditionally known as polarity constraints are not that “special”, i.e.

they need not be encoded in a specific polarity module in the grammar-- a welcome result, as it

makes polarity less anomalous. This point is worth emphasizing, because it has been ignored in

most of the downward-entailingness based literature, which tacitly assumed the existence of

such a module, like e.g. the Binding Theory module, which applied as global filters on LFs or

semantic representations. Admitting such a module is highly unsatisfactory since it in effect just

stipulates licensing conditions as composition-external global constraints. Ultimately we would

like to be able to derive the distribution of a given polarity item from the tools and principles that

are independently needed in the grammar, and dispense entirely with polarity-specific filters of

the traditional kind.

Finally, since free choice is not equivalent to universal quantification, we can no longer

maintain the ambiguity thesis for English any as one in terms of quantificational force. I have

provided an alternative for a uniform indefinite any within the (non)veridicality hypothesis

which I hope to have shown to fare better than accounts which postulate two quantificationally

distinct anys. Based on the fact that the distinction between free choice items and affective

polarity items is indeed lexicalized crosslinguistically, one can of course still argue that there are

two indefinite anys, one free choice item and one affective polarity item. This option is not

excluded by the analysis offered here, but whether to adopt it or not seems to have been reduced

to a relatively harmless, and perhaps even trivial, terminological question.

65

Acknowledgments

I am grateful to Larry Horn and Josep Quer for many exciting and lengthy discussions on free

choice which helped me tremendously in forming my ideas about the topic. I am also most

thankful to Jason Merchant for his detailed comments on this manuscript which led to

substantial improvements in both content and form, and to the two anonymous L&P reviewers

whose comments I benefitted greatly from. Thanks also to Bart Geurts, Jack Hoeksema, Alice

Ter Meulen, and Kjell-Johan Sæbø and Lucia Tovena for discussion, and to the following

people for their judgements and help with their native languages: Gregory Ward and Jason

Merchant for English; Josep Quer for Catalan and Spanish; and Danièle Godard for her

generous comments and help with the French data. Material based on this paper was presented

at the Colloque de Syntaxe et Sémantique de Paris 1997, the 11th Amsterdam Colloquium

(Giannakidou 1997b), the Universities of Nijmegen, Utrecht and Northwestern, and the 4th

International Conference on Greek Linguistics held at the University of Cyprus in 1999. Many

thanks to the audiences of those events for their feedback. The research reported here is

supported by the Royal Dutch Academy of Sciences (KNAW), whose contribution is hereby

gratefully acknowledged.

ReferencesBerman, Steve: 1991, On the Semantics and Logical Form of Wh-clauses, PhD dissertation,

University of Massachussets, Amherst.

Bolinger, Dwight: 1960, ‘Linguistic science and linguistic engineering’, Word 16, 374-91.

Bolinger, Dwight: 1977, Meaning and Form. Longman, London. (Chapter 2, Any and some.)

Bosque, Ignacio: 1996, ‘La polaridad modal’, Ms., University of Madrid.

Chierchia, Gennaro: 1995, ‘Individual-level predicates as inherent generics’, in G. Carlson and G.

Pelletier (eds.), The Generic Book., Chicago University Press, Chicago, pp. 176-224.

Chierchia, Gennaro, and Sally McConell-Ginet: 1991, Meaning and Grammar. MIT Press,

Cambridge, MA.

McCawley, James D.: 1981, Everything that Linguists have Always Wanted to Know about

Logic (but were ashamed to ask), University of Chicago Press, Chicago.

Dahl, Östen: 1999, Review of Indefinite Pronouns by Martin Haspelmath. Linguistics and

Philosophy 22: 663-678.

Davison, Alice: 1980, ‘Any as universal or existential’, in J. van der Auwera (ed.), The

Semantics of Determiners, Croom Helm, London, pp. 11-34.

Dayal, Veneeta: 1995, ‘Licensing any in non-negative/non-modal contexts’. in the Proceedings

of Semantics and Linguistic Theory (SALT) V, pp. 72-93.

Dayal, Veneeta: 1997, ‘Free choice and -ever: identity and free choice readings’, in the

66

Proceedings of Semantics and Linguistic Theory (SALT) VII.

Dayal, Veneeta: 1998, ‘Any as inherently modal’, Linguistics and Philosophy 21: 433-76.

den Dikken, Marcel, and Anastasia Giannakidou: 2001, ‘Aggressively non-D-linked phrases as

polarity items’ Ms., CUNY and University of Groningen.

Farkas, Donka F.: 1992, ‘On the semantics of subjunctive complements’, in P. Hirschbühler

and K. Körner (eds.) Romance Languages and Modern Linguistic Theory, pp. 69-104.

Amsterdam, John Benjamins.

Von Fintel, Kai: 1994, Restrictions on Quantifier Domains, PhD dissertation, University of

Massachussets, Amherst.

Von Fintel, Kai: 1999, ‘NPI-Licensing, Strawson-Entailment, and Context-Dependency’,

Journal of Semantics 16: 97-148.

Von Fintel, Kai: 2000, ‘Whatever’, In the Proceedings of Semantics and Linguistic Theory

(SALT) X.

Geurts, Bart: 1999, Presupposition and Pronouns. Elsevier, Oxford.

Giannakidou, Anastasia: 1995, ‘Subjunctive, habituality and negative polarity’, in the

Proceedings of Semantics and Linguistic Theory (SALT) V, pp. 132-150.

Giannakidou, Anastasia: 1997a, The Landscape of Polarity Items, PhD dissertation, University

of Groningen.

Giannakidou, Anastasia. 1997b, ‘Linking sensitivity to limited distribution: the case of free

choice’, In P. Dekker et al. (eds), Proceedings of the 11th Amsterdam Colloquium.

ILLC, University of Amsterdam. 139-145.

Giannakidou, Anastasia: 1998, Polarity Sensitivity as (Non)veridical Dependency, John

Benjamins, Amsterdam.

Giannakidou, Anastasia: 1999, ‘Affective dependencies’, Linguistics and Philosophy 22, 367-

421.

Giannakidou, Anastasia: 2000, ‘Negative …Concord?’, Natural Language and Linguistic

Theory 18, 457-523.

Giannakidou, Anastasia and Frans Zwarts: 1999, ‘Temporal, aspectual operators and

(Non)Veridicality’. To appear in The Syntax and Semantics of Tense and Mood

Selection, ed.by A. Giorgi, J. Higginbotham, and F. Pianesi, Oxford University Press.

Groenendijk , Jeroen and Martin Stokhof: 1984, Studies in the semantics of questions and

the pragmatics of answers, PhD dissertation, University of Amsterdam.

Haspelmath, Martin: 1997, Indefinite Pronouns. Oxford: Clarendon.

Heim, Irene: 1982, The Semantics of Definite and Indefinite Noun Phrases. PhD dissertation,

University of Massachussets, Amherst.

Hintikka, Jaako: 1962, Knowledge and Belief: an Introduction to the Logic of the Two Notions.

Cornell University Press, Ithaca, NY.

67

Hoeksema, Jacob: 1990, ‘Exploring exception phrases’, in the Proceedings of the 7th

Amsterdam Colloquium, ed. By Martin Stokhof and Leen Torenvliet, pp. 165-190,

Mathematisch centrum, Amsterdam.

Horn, Laurence R: 1972, On the Semantic Properties of Logical Operators in English, PhD

dissertation, University of California Los Angeles.

Horn, Laurence R.: 1999, ‘Any and (-)ever. Free choice and free relatives’. To appear in

Proceedings of IATL [Israeli Association for Theoretical Linguistics] 15.

Horn, Laurence R.: 2000, ‘Pick a theory: not just any theory’, in L. Horn & Y. Kato (eds.),

Negation and Polarity: Syntactic and Semantic Perspectives, Oxford University Press,

Oxford, pp. 147-192.

Horn, Laurence R. and Y.-S. Lee: 1995, ‘Diagnosing the diagnostics’, Ms., Yale University.

Israel, Michael: 1996, ‘Polarity sensitivity as lexical semantics’, Linguistics and Philosophy 19,

619-666.

Jacobson, Pauline: 1995, ‘The quantificational force of English free relatives’, In Bach et al.

(eds.) Quantification in Natural Language, Kluwer, Dordrecht, pp. 451-486.

Kadmon, Nirit and Fred Landman: 1993, ‘Any’, Linguistics and Philosophy 16, 353-422

Kamp, Hans: 1978, ‘Semantics versus pragmatics’, in Guenthner, F. and S. J. Schmidt (eds.),

Formal Semantics and Pragmatics, Reidel, Dordrecht, pp. 255-287.

Karttunen, Lauri: 1977, ‘Syntax and semantics of questions’, Linguistics and Philosophy 1, 3-44.

Kiparksy, Paul, and Carol Kiparsky: 1971, ‘Fact’, in D. Steinberg and L. Jacobovits (eds),

Semantics, Cambridge University Press, pp. 345-369.

Kratzer, Angelika: 1981, ‘On the notional category of modality’, in H. Eikmeyer and H. Rieser

(eds), Words, Worlds, and Contexts, de Gruyter, Berlin, pp. 38-74.

Kratzer, Angelika: 1995, ‘Stage-level and individual-level predicates’, in G. Carlson & F. J.

Pelletier (eds.), The Generic Book, Chicago University Press, Chicago, pp. 125-176.

Krifka, Manfred, F. J. Pelletier, Greg Carlson, Alice ter Meulen, Gennaro Chierchia &

Godehard Link: 1995, ‘Genericity: an introduction’, in G. Carlson & F. J. Pelletier

(eds.), The Generic Book, University of Chicago Press, Chicago, pp. 1-124.

Ladusaw, William A.: 1979, Polarity Sensitivity as Inherent Scope Relations, PhD dissertation,

University of Texas at Austin.

Ladusaw, William A.: 1986, ‘Principles of semantic filtering’, in the Proceedings of the West

Coast Conference on Formal Linguistics (WCCFL) 5.

Ladusaw, William A.: 1996, ‘Negation and polarity items’, in S. Lappin (ed.) The Handbook of

Contemporary Semantic Theory, Blackwell, Oxford. pp.321-343.

Lee, Chungmin: 1997, ‘Negative polarity and free choice: where do they come from?’ In P.

Dekker et al. (eds), Proceedings of the 11th Amsterdam Colloquium. ILLC, University of

Amsterdam, pp. 217-222.

68

Lee, Young-Suk and Laurence R. Horn: 1994, ‘Any as an indefinite plus even’, Ms., Yale

University.

LeGrand, Jean Ehrenkranz: 1975, Or and Any: The semantics and syntax of Two logical

Operators, PhD dissertation, University of Chicago.

Lin, Jo-Wang: 1996, Polarity Licensing and Wh-phrase Quantification in Chinese, PhD

dissertation, University of Massachussets, Amherst.

Linebarger, Marcia: 1980, The Grammar of Negative Polarity, PhD dissertation, MIT.

Linebarger, Marcia: 1987, ‘Negative polarity and grammatical representation’, Linguistics

and Philosophy 10, 325-387.

Merchant, Jason: 2000, ‘Antecedent contained deletion in negative polarity items’, Syntax 3,

134-150.

Quer, Josep: 1998, Mood at the Interface, PhD dissertation, University of Utrecht.

Quer, Josep: 1999, ‘The quantificational force of free choice items’, Ms., University of

Amsterdam.

Quer, Josep: 2000, ‘Licensing free choice items in hostile environments: the role of aspect and

mood’, Ms., University of Amsterdam.

Quine, W. van Orman: 1960, Word and Object, MIT Press, Cambridge, Mass.

Partee, Barbara: 1987, ‘Noun phrase interpretation and type-shifting principles’, in J.

Groenendijk et al. (eds), Studies in Discourse Representation Theory and the Theory of

Generalized Quantifiers, Foris, Dordrecht, pp. 115-143.

Progovac, Liljiana: 1994, Negative and Positive Polarity, Cambridge University Press.

Reichenbach, Hans: 1947, Elements of Symbolic Logic, The Free Press, New York.

Rullmann, Hotze: 1996, Two Types of Negative Polarity Items’, in the Proceedings of

Northeastern Linguistics Society (NELS) 26.

Sæbø, Kjell-Johan: this issue, ‘The semantics of Scandinavian free choice items’.

Tovena, Lucia: 1999, The Fine Structure of Polarity Items, Garland, New York.

Tovena, Lucia and Jacques Jayez: 1997, ‘The modal arbitrriness of any’, Ms., University of Geneva and

EHESS, Paris.

Tredinnick, Victoria: 1996, ‘On the interpretation and distribution of –ever in English free relatives’,

Proceedings of CONSOLE 4.

Vikner, Carl: 1999, ‘Episodic and habitual temporal connectives in Danish, Norwegian and

Swedish’, NORDSEM Report, University of Copenhagen.

Zwarts, Frans: 1986, Categoriale Grammatica en Algebraische Semantiek. Een studie naar

negatie en polariteit in het Nederlands, PhD dissertation, University of Groningen.

Zwarts, Frans: 1995, ‘Nonveridical contexts’, Linguistic Analysis 25, 286-312.